Silver iodide crystals have long been used to “seed” clouds and trigger precipitation, but scientists have never been entirely sure why the material works so well for that purpose. Researchers at TU Wien in Austria are now a step closer to solving the mystery thanks to a new study that characterized surfaces of the material in atomic-scale detail.
“Silver iodide has been used in atmospheric weather modification programs around the world for several decades,” explains Jan Balajka from TU Wien’s Institute of Applied Physics, who led this research. “In fact, it was chosen for this purpose as far back as the 1940s because of its atomic crystal structure, which is nearly identical to that of ice – it has the same hexagonal symmetry and very similar distances between atoms in its lattice structure.”
The basic idea, Balajka continues, originated with the 20th-century American atmospheric scientist Bernard Vonnegut, who suggested in 1947 that introducing small silver iodide (AgI) crystals into a cloud could provide nuclei for ice to grow on. But while Vonnegut’s proposal worked (and helped to inspire his brother Kurt’s novel Cat’s Cradle), this simple picture is not entirely accurate. The stumbling block is that nucleation occurs at the surface of a crystal, not inside it, and the atomic structure of an AgI surface differs significantly from its interior.
A task that surface science has solved
To investigate further, Balajka and colleagues used high-resolution atomic force microscopy (AFM) and advanced computer simulations to study the atomic structure of 2‒3 nm diameter AgI crystals when they are broken into two pieces. The team’s measurements revealed that the surfaces of both freshly cleaved structures differed from those found inside the crystal.
More specifically, team member Johanna Hütner, who performed the experiments, explains that when an AgI crystal is cleaved, the silver atoms end up on one side while the iodine atoms appear on the other. This has implications for ice growth, because while the silver side maintains a hexagonal arrangement that provides an ideal template for the growth of ice layers, the iodine side reconstructs into a rectangular pattern that no longer lattice-matches the hexagonal symmetry of ice crystals. The iodine side is therefore incompatible with the epitaxial growth of hexagonal ice.
“Our works solves this decades-long controversy of the surface vs bulk structure of AgI, and shows that structural compatibility does matter,” Balajka says.
Difficult experiments
According to Balajka, the team’s experiments were far from easy. Many experimental methods for studying the structure and properties of material surfaces are based on interactions with charged particles such as electrons or ions, but AgI is an electrical insulator, which “excludes most of the tools available,” he explains. Using AFM enabled them to overcome this problem, he adds, because this technique detects interatomic forces between a sharp tip and the surface and does not require a conductive sample.
Another problem is that AgI is photosensitive and decomposes when exposed to visible light. While this property is useful in other contexts – AgI was a common ingredient in early photographic plates – it created complications for the TU Wien team. “Conventional AFM setups make use of optical laser detection to map the topography of a sample,” Balajka notes.
To avoid destroying their sample while studying it, the researchers therefore had to use a non-contact AFM based on a piezoelectric sensor that detects electrical signals and does not require optical readout. They also adapted their setup to operate in near-darkness, using only red light while manipulating the Ag to ensure that stray light did not degrade the samples.
The computational modelling part of the work introduced yet another hurdle to overcome. “Both Ag and I are atoms with a high number of electrons in their electron shells and are thus highly polarizable,” Balajka explains. “The interaction between such atoms cannot be accurately described by standard computational modelling methods such as density functional theory (DFT), so we had to employ highly accurate random-phase approximation (RPA) calculations to obtain reliable results.”
Highly controlled conditions
The researchers acknowledge that their study, which is detailed in Science Advances, was conducted under highly controlled conditions – ultrahigh vacuum, low pressure and temperature and a dark environment – that are very different from those that prevail inside real clouds. “The next logical step for us is therefore to confirm whether our findings hold under more representative conditions,” Balajka says. “We would like to find out whether the structure of AgI surfaces is the same in air and water, and if not, why.”
The researchers would also like to better understand the atomic arrangement of the rectangular reconstruction of the iodine surface. “This would complete the picture for the use of AgI in ice nucleation, as well as our understanding of AgI as a material overall,” Balajka says.
Using a novel spectroscopy technique, physicists in Japan have revealed how organic materials accumulate electrical charge through long-term illumination by sunlight – leading to material degradation. Ryota Kabe and colleagues at the Okinawa Institute of Science and Technology have shown how charge separation occurs gradually via a rare multi-photon ionization process, offering new insights into how plastics and organic semiconductors degrade in sunlight.
In a typical organic solar cell, an electron-donating material is interfaced with an electron acceptor. When the donor absorbs a photon, one of its electrons may jump across the interface, creating a bound electron-hole pair which may eventually dissociate – creating two free charges from which useful electrical work can be extracted.
Although such an interface vastly boosts the efficiency of this process, it is not necessary for charge separation to occur when an electron donor is illuminated. “Even single-component materials can generate tiny amounts of charge via multiphoton ionization,” Kabe explains. “However, experimental evidence has been scarce due to the extremely low probability of this process.”
To trigger charge separation in this way, an electron needs to absorb one or more additional photons while in its excited state. Since the vast majority of electrons fall back into their ground states before this can happen, the spectroscopic signature of this charge separation is very weak. This makes it incredibly difficult to detect using conventional spectroscopy techniques, which can generally only make observations over timescales of up to a few milliseconds.
The opposite approach
“While weak multiphoton pathways are easily buried under much stronger excited-state signals, we took the opposite approach in our work,” Kabe describes. “We excited samples for long durations and searched for traces of accumulated charges in the slow emission decay.”
Key to this approach was an electron donor called NPD. This organic material has a relatively long triplet lifetime, where an excited electron is prevented from transitioning back to its ground state. As a result, these molecules emit phosphorescence over relatively long timescales.
In addition, Kabe’s team dispersed their NPD samples into different host materials with carefully selected energy levels. In one medium, the energies of both the highest-occupied and lowest-unoccupied molecular orbitals lay below NPD’s corresponding levels, so that the host material acted as an electron acceptor. As a result, charge transfer occurred in the same way as it would across a typical donor-acceptor interface.
Yet in another medium, the host’s lowest-unoccupied orbital lay above NPD’s – blocking charge transfer, and allowing triplet states to accumulate instead. In this case, the only way for charge separation to occur was through multi-photon ionization.
Slow emission decay analysis
Since NPD’s long triplet lifetime allowed its electrons to be excited gradually over an extended period of illumination, its weak charge accumulation became detectable through slow emission decay analysis. In contrast, more conventional methods involve multiple, ultra-fast laser pulses, severely restricting the timescale over which measurements can be made. Altogether, this approach enabled the team to clearly distinguish between the two charge generation pathways.
“Using this method, we confirmed that charge generation occurred via resonance-enhanced multiphoton ionization mediated by long-lived triplet states, even in single-component organic materials,” Kabe describes.
This result offers insights into how plastics and organic semiconductors are degraded by sunlight over years or decades. The conventional explanation is that sunlight generates free radicals. These are molecules that lose an electron through ionization, leaving behind an unpaired electron which readily reacts with other molecules in the surrounding environment. Since photodegradation unfolds over such a long timescale, researchers could not observe this charge generation in single-component organic materials – until now.
“The method will be useful for analysing charge behaviour in organic semiconductor devices and for understanding long-term processes such as photodegradation that occur gradually under continuous light exposure,” Kabe says.
Fermilab has officially opened a new building named after the particle physicist Helen Edwards. Officials from the lab and the US Department of Energy (DOE) opened the Helen Edwards Engineering Research Center at a ceremony held on 5 December. The new building is the lab’s largest purpose-built lab and office space since the lab’s iconic Wilson Hall, which was completed in 1974.
Construction of the Helen Edwards Engineering Research Center began in 2019 and was completed three years later. The centre is an 7500 m2 multi-story lab and office building that is adjacent and connected to Wilson Hall.
The new centre is designed as a collaborative lab where engineers, scientists and technicians design, build and test technologies across several areas of research such as neutrino science, particle detectors, quantum science and electronics.
The centre also features cleanrooms, vibration-sensitive labs and cryogenic facilities in which the components of the near detector for the Deep Underground Neutrino Experiment will be assembled and tested.
A pioneering spirit
With a PhD in experimental particle physics from Cornell University, Edwards was heavily involved with commissioning the university’s 10 GeV electron synchrotron. In 1970 Fermilab’s director Robert Wilson appointed Edwards as associate head of the lab’s booster section and she later became head of the accelerator division.
While at Fermilab, Edwards’ primary responsibility was designing, constructing, commissioning and operating the Tevatron, which led to the discoveries of the top quark in 1995 and the tau neutrino in 2000.
Edwards retired in the early 1990s but continued to work as guest scientists at Fermilab and officially switched the Tevatron off during a ceremony held on 30 September 2011. Edwards died in 2016.
Darío Gil, the undersecretary for science at the DOE says that Edwards’ scientific work “is a symbol of the pioneering spirit of US research”.
“Her contributions to the Tevatron and the lab helped the US become a world leader in the study of elementary particles,” notes Gil. “We honour her legacy by naming this research centre after her as Fermilab continues shaping the next generation of research using [artificial intelligence], [machine learning] and quantum physics.”
A proposed new way of defining the standard unit of electrical resistance would do away with the need for strong magnetic fields when measuring it. The new technique is based on memristors, which are programmable resistors originally developed as building blocks for novel computing architectures, and its developers say it would considerably simplify the experimental apparatus required to measure a single quantum of resistance for some applications.
Electrical resistance is a physical quantity that represents how much a material opposes the flow of electrical current. It is measured in ohms (Ω), and since 2019, when the base units of the International System of Units (SI) were most recently revised, the ohm has been defined in terms of the von Klitzing constant h/e2, where h and e are the Planck constant and the charge on an electron, respectively.
To measure this resistance with high precision, scientists use the fact that the von Klitzing constant is related to the quantized change in the Hall resistance of a two-dimensional electron system (such as the one that forms in a semiconductor heterostructure) in the presence of a strong magnetic field. This quantized change in resistance is known as the quantum Hall effect (QHE), and in a material like GaAs or AlGaAs, it shows up at fields of around 10 Tesla. Generating such high fields typically requires a superconducting electromagnet, however.
A completely different approach
Researchers connected to a European project called MEMQuD are now advocating a completely different approach. Their idea is based on memristors, which are programmable resistors that “remember” their previous resistance state even after they have been switched off. This previous resistance state can be changed by applying a voltage or current.
The MEMQuD team reports that the quantum conductance levels achieved in this set-up are precise enough to be exploited as intrinsic standard values. Indeed, a large inter-laboratory comparison confirmed that the values deviated by just -3.8% and 0.6% from the agreed SI values for the fundamental quantum of conductance, G0, and 2G0, respectively. The researchers attribute this precision to tight, atomic-level control over the morphology of the nanochannels responsible for quantum conductance effects, which they achieved by electrochemically polishing the silver filaments into the desired configuration.
A national metrology institute condensed into a microchip
The researchers say their results are building towards a concept known as an “NMI-in-a-chip” – that is, condensing the services of a national metrology institute into a microchip. “This could lead to measuring devices that have their resistance references built-in directly into the chip,” says Milano, “so doing away with complex measurements in laboratories and allowing for devices with zero-chain traceability – that is, those that do not require calibration since they have embedded intrinsic standards.”
“Notably, this method can be demonstrated at room temperature and under ambient conditions, in contrast to conventional methods that require cryogenic and vacuum equipment, which is expensive and require a lot of electrical power,” Okazaki says. “If such a user-friendly quantum standard becomes more stable and its uncertainty is improved, it could lead to a new calibration scheme for ensuring the accuracy of electronics used in extreme environments, such as space or the deep ocean, where traditional quantum standards that rely on cryogenic and vacuum conditions cannot be readily used.”
The MEMQuD researchers, who report their work in Nature Nanotechnology, now plan to explore ways to further decrease deviations from the agreed SI values for G0 and 2G0. These include better material engineering, an improved measurement protocol, and strategies for topologically protecting the memristor’s resistance.
Travis Humble is a research leader who’s thinking big, dreaming bold, yet laser-focused on operational delivery. The long-game? To translate advances in fundamental quantum science into a portfolio of enabling technologies that will fast-track the practical deployment of quantum computers for at-scale scientific, industrial and commercial applications.
Validation came in spades last month when, despite the current turbulence around US science funding, QSC was given follow-on DOE backing of $125 million over five years (2025–30) to create “a new scientific ecosystem” for fault-tolerant, quantum-accelerated high-performance computing (QHPC). In short, QSC will target the critical research needed to amplify the impact of quantum computing through its convergence with leadership-class exascale HPC systems.
“Our priority in Phase II QSC is the creation of a common software ecosystem to host the compilers, programming libraries, simulators and debuggers needed to develop hybrid-aware algorithms and applications for QHPC,” explains Humble. Equally important, QSC researchers will develop and integrate new techniques in quantum error correction, fault-tolerant computing protocols and hybrid algorithms that combine leading-edge computing capabilities for pre- and post-processing of quantum programs. “These advances will optimize quantum circuit constructions and accelerate the most challenging computational tasks within scientific simulations,” Humble adds.
Classical computing, quantum opportunity
At the heart of the QSC programme sits ORNL’s leading-edge research infrastructure for classical HPC, a capability that includes Frontier, the first supercomputer to break the exascale barrier and still one of the world’s most powerful. On that foundation, QSC is committed to building QHPC architectures that take advantage of both quantum computers and exascale supercomputing to tackle all manner of scientific and industrial problems beyond the reach of today’s HPC systems alone.
“Hybrid classical-quantum computing systems are the future,” says Humble. “With quantum computers connecting both physically and logically to existing HPC systems, we can forge a scalable path to integrate quantum technologies into our scientific infrastructure.”
Quantum acceleration ORNL’s current supercomputer, Frontier, was the first high-performance machine to break the exascale barrier. Plans are in motion for a next-generation supercomputer, Discovery, to come online at ORNL by 2028. (Courtesy: Carlos Jones/ORNL, US DOE)
Industry partnerships are especially important in this regard. Working in collaboration with the likes of IonQ, Infleqtion and QuEra, QSC scientists are translating a range of computationally intensive scientific problems – quantum simulations of exotic matter, for example – onto the vendors’ quantum computing platforms, generating excellent results out the other side.
“With our broad representation of industry partners,” notes Humble, “we will establish a common framework by which scientific end-users, software developers and hardware architects can collaboratively advance these tightly coupled, scalable hybrid computing systems.”
It’s a co-development model that industry values greatly. “Reciprocity is key,” Humble adds. “At QSC, we get to validate that QHPC can address real-world research problems, while our industry partners gather user feedback to inform the ongoing design and optimization of their quantum hardware and software.”
Quantum impact
Innovation being what it is, quantum computing systems will continue to trend on an accelerating trajectory, with more qubits, enhanced fidelity, error correction and fault-tolerance key reference points on the development roadmap. Phase II QSC, for its part, will integrate five parallel research thrusts to advance the viability and uptake of QHPC technologies.
The collaborative software effort, led by ORNL’s Vicente Leyton, will develop openQSE, an adaptive, end-to-end software ecosystem for QHPC systems and applications. Yigit Subasi from Los Alamos National Laboratory (LANL) will lead the hybrid algorithms thrust, which will design algorithms that combine conventional and quantum methods to solve challenging problems in the simulation of model materials.
Meanwhile, the QHPC architectures thrust, under the guidance of ORNL’s Chris Zimmer, will co-design hybrid computing systems that integrate quantum computers with leading-edge HPC systems. The scientific applications thrust, led by LANL’s Andrew Sornberger, will develop and validate applications of quantum simulation to be implemented on prototype QHPC systems. Finally, ORNL’s Michael McGuire will lead the thrust to establish experimental baselines for quantum materials that ultimately validate QHPC simulations against real-world measurements.
Longer term, ORNL is well placed to scale up the QHPC model. After all, the laboratory is credited with pioneering the hybrid supercomputing model that uses graphics processing units in addition to conventional central processing units (including the launch in 2012 of Titan, the first supercomputer of this type operating at over 10 petaFLOPS).
“The priority for all the QSC partners,” notes Humble, “is to transition from this still-speculative research phase in quantum computing, while orchestrating the inevitable convergence between quantum technology, existing HPC capabilities and evolving scientific workflows.”
Collaborate, coordinate, communicate
Much like its NQISRC counterparts (which have also been allocated further DOE funding through 2030), QSC provides the “operational umbrella” for a broad-scope collaboration of more than 300 scientists and engineers from 20 partner institutions. With its own distinct set of research priorities, that collective activity cuts across other National Laboratories (Los Alamos and Pacific Northwest), universities (among them Berkeley, Cornell and Purdue) and businesses (including IBM and IQM) to chart an ambitious R&D pathway addressing quantum-state (qubit) resilience, controllability and, ultimately, the scalability of quantum technologies.
“QSC is a multidisciplinary melting pot,” explains Humble, “and I would say, alongside all our scientific and engineering talent, it’s the pooled user facilities that we are able to exploit here at Oak Ridge and across our network of partners that gives us our ‘grand capability’ in quantum science [see box, “Unique user facilities unlock QSC opportunities”]. Certainly, when you have a common research infrastructure, orchestrated as part a unified initiative like QSC, then you can deliver powerful science that translates into real-world impacts.”
Unique user facilities unlock QSC opportunities
Neutron insights ORNL director Stephen Streiffer tours the linear accelerator tunnel at the Spallation Neutron Source (SNS). QSC scientists are using the SNS to investigate entirely new classes of strongly correlated materials that demonstrate topological order and quantum entanglement. (Courtesy: Alonda Hines/ORNL, US DOE)
Deconstructed, QSC’s Phase I remit (2020–25) spanned three dovetailing and cross-disciplinary research pathways: discovery and development of advanced materials for topological quantum computing (in which quantum information is stored in a stable topological state – or phase – of a physical system rather than the properties of individual particles or atoms); development of next-generation quantum sensors (to characterize topological states and support the search for dark matter); as well as quantum algorithms and simulations (for studies in fundamental physics and quantum chemistry).
Underpinning that collective effort: ORNL’s unique array of scientific user facilities. A case in point is the Spallation Neutron Source (SNS), an accelerator-based neutron-scattering facility that enables a diverse programme of pure and applied research in the physical sciences, life sciences and engineering. QSC scientists, for example, are using SNS to investigate entirely new classes of strongly correlated materials that demonstrate topological order and quantum entanglement – properties that show great promise for quantum computing and quantum metrology applications.
“The high-brightness neutrons at SNS give us access to this remarkable capability for materials characterization,” says Humble. “Using the SNS neutron beams, we can probe exotic materials, recover the neutrons that scatter off of them and, from the resultant signals, infer whether or not the materials exhibit quantum properties such as entanglement.”
While SNS may be ORNL’s “big-ticket” user facility, the laboratory is also home to another high-end resource for quantum studies: the Center for Nanophase Material Science (CNMS), one of the DOE’s five national Nanoscience Research Centers, which offers QSC scientists access to specialist expertise and equipment for nanomaterials synthesis; materials and device characterization; as well as theory, modelling and simulation in nanoscale science and technology.
Thanks to these co-located capabilities, QSC scientists pioneered another intriguing line of enquiry – one that will now be taken forward elsewhere within ORNL – by harnessing so-called quantum spin liquids, in which electron spins can become entangled with each other to demonstrate correlations over very large distances (relative to the size of individual atoms).
In this way, it is possible to take materials that have been certified as quantum-entangled and use them to design new types of quantum devices with unique geometries – as well as connections to electrodes and other types of control systems – to unlock novel physics and exotic quantum behaviours. The long-term goal? Translation of quantum spin liquids into a novel qubit technology to store and process quantum information.
SNS, CNMS and Oak Ridge Leadership Computing Facility (OLCF) are DOE Office of Science user facilities.
When he’s not overseeing the technical direction of QSC, Humble is acutely attuned to the need for sustained and accessible messaging. The priority? To connect researchers across the collaboration – physicists, chemists, material scientists, quantum information scientists and engineers – as well as key external stakeholders within the DOE, government and industry.
“In my experience,” he concludes, ”the ability of the QSC teams to communicate efficiently – to understand each other’s concepts and reasoning and to translate back and forth across disciplinary boundaries – remains fundamental to the success of our scientific endeavours.”
The next generation Quantum science graduate students and postdoctoral researchers present and discuss their work during a poster session at the fifth annual QSC Summer School. Hosted at Purdue University in April this year, the school is one of several workforce development efforts supported by QSC. (Courtesy: Dave Mason/Purdue University)
With an acknowledged shortage of skilled workers across the quantum supply chain, QSC is doing its bit to bolster the scientific and industrial workforce. Front-and-centre: the fifth annual QSC Summer School, which was held at Purdue University in April this year, hosting 130 graduate students (the largest cohort to date) through an intensive four-day training programme.
The Summer School sits as part of a long-term QSC initiative to equip ambitious individuals with the specialist domain knowledge and skills needed to thrive in a quantum sector brimming with opportunity – whether that’s in scientific research or out in industry with hardware companies, software companies or, ultimately, the end-users of quantum technologies in key verticals like pharmaceuticals, finance and healthcare.
“While PhD students and postdocs are integral to the QSC research effort, the Summer School exposes them to the fundamental ideas of quantum science elaborated by leading experts in the field,” notes Vivien Zapf, a condensed-matter physicist at Los Alamos National Laboratory who heads up QSC’s advanced characterization efforts.
“It’s all about encouraging the collective conversation,” she adds, “with lots of opportunities for questions and knowledge exchange. Overall, our emphasis is very much on training up scientists and engineers to work across the diversity of disciplines needed to translate quantum technologies out of the lab into practical applications.”
The programme isn’t for the faint-hearted, though. Student delegates kicked off this year’s proceedings with a half-day of introductory presentations on quantum materials, devices and algorithms. Next up: three and a half days of intensive lectures, panel discussions and poster sessions covering everything from entangled quantum networks to quantum simulations of superconducting qubits.
Many of the Summer School’s sessions were also made available virtually on Purdue’s Quantum Coffeehouse Live Stream on YouTube – the streamed content reaching quantum learners across the US and further afield. Lecturers were drawn from the US National Laboratories, leading universities (such as Harvard and Northwestern) and the quantum technology sector (including experts from IBM, PsiQuantum, NVIDIA and JPMorganChase).
As a physicist in industry, I spend my days developing new types of photovoltaic (PV) panels. But I’m also keen to do something for the transition to green energy outside work, which is why I recently installed two PV panels on the balcony of my flat in Munich. Fitting them was great fun – and I can now enjoy sunny days even more knowing that each panel is generating electricity.
However, the panels, which each have a peak power of 440 W, don’t cover all my electricity needs, which prompted me to take an interest in a plan to build six wind turbines in a forest near me on the outskirts of Munich. Curious about the project, I particularly wanted to find out when the turbines will start generating electricity for the grid. So when I heard that a weekend cycle tour of the site was being organized to showcase it to local residents, I grabbed my bike and joined in.
As we cycle, I discover that the project – located in Forstenrieder Park – is the joint effort of four local councils and two “citizen-energy” groups, who’ve worked together for the last five years to plan and start building the six turbines. Each tower will be 166 m high and the rotor blades will be 80 m long, with the plan being for them to start operating in 2027.
I’ve never thought of Munich as a particularly windy city, but at the height at which the blades operate, there’s always a steady, reliable flow of wind
I’ve never thought of Munich as a particularly windy city. But tour leader Dieter Maier, who’s a climate adviser to Neuried council, explains that at the height at which the blades operate, there’s always a steady, reliable flow of wind. In fact, each turbine has a designed power output of 6.5 MW and will deliver a total of 10 GWh in energy over the course of a year.
Practical questions
Cycling around, I’m excited to think that a single turbine could end up providing the entire electricity demand for Neuried. But installing wind turbines involves much more than just the technicalities of generating electricity. How do you connect the turbines to the grid? How do you ensure planes don’t fly into the turbines? What about wildlife conservation and biodiversity?
At one point of our tour, we cycle round a 90-degree bend in the forest and I wonder how a huge, 80 m-long blade will be transported round that kind of tight angle? Trees will almost certainly have to be felled to get the blade in place, which sounds questionable for a supposedly green project. Fortunately, project leaders have been working with the local forest manager and conservationists, finding ways to help improve the local biodiversity despite the loss of trees.
As a representative of BUND (one of Germany’s biggest conservation charities) explains on the tour, a natural, or “unmanaged”, forest consists of a mix of areas with a higher or lower density of trees. But Forstenrieder Park has been a managed forest for well over a century and is mostly thick with trees. Clearing trees for the turbines will therefore allow conservationists to grow more of the bushes and plants that currently struggle to find space to flourish.
Cut and cover Trees in Forstenrieder Park have had to be chopped down to provide room for new wind turbines to be installed, but the open space will let conservationists grow plants and bushes to boost biodiversity. (Courtesy: Janina Moereke)
To avoid endangering birds and bats native to this forest, meanwhile, the turbines will be turned off when the animals are most active, which coincidentally corresponds to low wind periods in Munich. Insurance costs have to be factored in too. Thankfully, it’s quite unlikely that a turbine will burn down or get ice all over its blades, which means liability insurance costs are low. But vandalism is an ever-present worry.
In fact, at the end of our bike tour, we’re taken to a local wind turbine that is already up and running about 13 km further south of Forstenrieder Park. This turbine, I’m disappointed to discover, was vandalized back in 2024, which led to it being fenced off and video surveillance cameras being installed.
But for all the difficulties, I’m excited by the prospect of the wind turbines supporting the local energy needs. I can’t wait for the day when I’m on my balcony, solar panels at my side, sipping a cup of tea made with water boiled by electricity generated by the rotor blades I can see turning round and round on the horizon.
Excess radiation Gamma-ray intensity map excluding components other than the halo, spanning approximately 100° in the direction of the centre of the Milky Way. The blank horizontal bar is the galactic plane area, which was excluded from the analysis to avoid strong astrophysical radiation. (Courtesy: Tomonori Totani/The University of Tokyo)
Gamma rays emitted from the halo of the Milky Way could be produced by hypothetical dark-matter particles. That is the conclusion of an astronomer in Japan who has analysed data from NASA’s Fermi Gamma-ray Space Telescope. The energy spectrum of the emission is what would be expected from the annihilation of particles called WIMPs. If this can be verified, it would mark the first observation of dark matter via electromagnetic radiation.
Since the 1930s astronomers have known that there is something odd about galaxies, galaxy clusters and larger structures in the universe. The problem is that there is not nearly enough visible matter in these objects to explain their dynamics and structure. A rotating galaxy, for example, should be flinging out its stars because it does not have enough self-gravitation to hold itself together.
Today, the most popular solution to this conundrum is the existence of a hypothetical substance called dark matter. Dark-matter particles would have mass and interact with each other and normal matter via the gravitational force, gluing rotating galaxies together. However, the fact that we have never observed dark matter directly means that the particles must rarely, if ever, interact via the other three forces.
Annihilating WIMPs
The weakly interacting massive particle (WIMP) is a dark-matter candidate that interacts via the weak nuclear force (or a similarly weak force). As a result of this interaction, pairs of WIMPs are expected to occasionally annihilate to create high-energy gamma rays and other particles. If this is true, dense areas of the universe such as galaxies should be sources of these gamma rays.
Now, Tomonori Totani of the University of Tokyo has analysed data from the Fermi telescope and identified an excess of gamma rays emanating from the halo of the Milky Way. What is more, Totani’s analysis suggests that the energy spectrum of the excess radiation (from about 10−100 GeV) is consistent with hypothetical WIMP annihilation processes.
“If this is correct, to the extent of my knowledge, it would mark the first time humanity has ‘seen’ dark matter,” says Totani. “This signifies a major development in astronomy and physics,” he adds.
While Totani is confident of his analysis, his conclusion must be verified independently. Furthermore, work will be needed to rule out conventional astrophysical sources of the excess radiation.
Catherine Heymans, who is Astronomer Royal for Scotland told Physics World, “I think it’s a really nice piece of work, and exactly what should be happening with the Fermi data”. The research is described in Journal of Cosmology and Astroparticle Physics. Heymans describes Totani’s paper as “well written and thorough”.
Researchers in the US have shed new light on the puzzling and complex flight physics of creatures such as hummingbirds, bumblebees and dragonflies that flap their wings to hover in place. According to an interdisciplinary team at the University of Cincinnati, the mechanism these animals deploy can be described by a very simple, computationally basic, stable and natural feedback mechanism that operates in real time. The work could aid the development of hovering robots, including those that could act as artificial pollinators for crops.
If you’ve ever watched a flapping insect or hummingbird hover in place – often while engaged in other activities such as feeding or even mating – you’ll appreciate how remarkable they are. To stay aloft and stable, these animals must constantly sense their position and motion and make corresponding adjustments to their wing flaps.
Feedback mechanism relies on two main components
Biophysicists have previously put forward many highly complex explanations for how they do this, but according to the Cincinnati team of Sameh Eisa and Ahmed Elgohary, some of this complexity is not necessary. Earlier this year, the pair developed their own mathematical and control theory based on a mechanism they call “extremum seeking for vibrational stabilization”.
Eisa describes this mechanism as “very natural” because it relies on just two main components. The first is the wing flapping motion itself, which he says is “naturally built in” for flapping creatures that use it to propel themselves. The second is a simple feedback mechanism involving sensations and measurements related to the altitude at which the creatures aim to stabilize their hovering.
The general principle, he continues, is that a system (in this case an insect or hummingbird) can steer itself towards a stable position by continuously adjusting a high-amplitude, high-frequency input control or signal (in this case, a flapping wing action). “This adjustment is simply based on the feedback of measurement (the insects’ perceptions) and stabilization (hovering) occurs when the system optimizes what it is measuring,” he says.
As well as being relatively easy to describe, Eisa tells Physics World that this mechanism is biologically plausible and computationally basic, dramatically simplifying the physics of hovering. “It is also categorically different from all available results and explanations in the literature for how stable hovering by insects and hummingbirds can be achieved,” he adds.
The researchers and colleagues. (Courtesy: S Eisa)
Interdisciplinary work
In the latest study, which is detailed in Physical Review E, the researchers compared their simulation results to reported biological data on a hummingbird and five flapping insects (a bumblebee, a cranefly, a dragonfly, a hawkmoth and a hoverfly). They found that their simulation fit the data very closely. They also ran an experiment on a flapping, light-sensing robot and observed that it behaved like a moth: it elevated itself to the level of the light source and then stabilized its hovering motion.
Eisa says he has always been fascinated by such optimized biological behaviours. “This is especially true for flyers, where mistakes in execution could potentially mean death,” he says. “The physics behind the way they do it is intriguing and it probably needs elegant and sophisticated mathematics to be described. However, the hovering creatures appear to be doing this very simply and I found discovering the secret of this puzzle very interesting and exciting.”
Eisa adds that this element of the work ended up being very interdisciplinary, and both his own PhD in applied mathematics and the aerospace engineering background of Elgohary came in very useful. “We also benefited from lengthy discussions with a biologist colleague who was a reviewer of our paper,” Eisa says. “Luckily, they recognized the value of our proposed technique and ended up providing us with very valuable inputs.”
Eisa thinks the work could open up new lines of research in several areas of science and engineering. “For example, it opens up new ideas in neuroscience and animal sensory mechanisms and could almost certainly be applied to the development of airborne robotics and perhaps even artificial pollinators,” he says. “The latter might come in useful in the future given the high rate of death many species of pollinating insects are encountering today.”
This episode of the Physics World Weekly podcast features Tim Hsieh of Canada’s Perimeter Institute for Theoretical Physics. We explore some of today’s hottest topics in quantum science and technology – including topological phases of matter; quantum error correction and quantum simulation.
Our conversation begins with an exploration of the quirky properties quantum matter and how these can be exploited to create quantum technologies. We look at the challenges that must be overcome to create large-scale quantum computers; and Hsieh reveals which problem he would solve first if he had access to a powerful quantum processor.
This interview was recorded earlier this autumn when I had the pleasure of visiting the Perimeter Institute and speaking to four physicists about their research. This is the third of those conversations to appear on the podcast.
Generative classification The CytoDiffusion classifier accurately identifies a wide range of blood cell appearances and detects unusual or rare blood cells that may indicate disease. The diagonal grid elements display original images of each cell type, while the off-diagonal elements show heat maps that provide insight into the model’s decision-making rationale. (Courtesy: Simon Deltadahl)
The shape and structure of blood cells provide vital indicators for diagnosis and management of blood disease and disorders. Recognizing subtle differences in the appearance of cells under a microscope, however, requires the skills of experts with years of training, motivating researchers to investigate whether artificial intelligence (AI) could help automate this onerous task. A UK-led research team has now developed a generative AI-based model, known as CytoDiffusion, that characterizes blood cell morphology with greater accuracy and reliability than human experts.
Conventional discriminative machine learning models can match human performance at classifying cells in blood samples into predefined classes. But discriminative models, which learn to recognise cell images based on expert labels, struggle with never-before-seen cell types and images from differing microscopes and staining techniques.
To address these shortfalls, the team – headed up at the University of Cambridge, University College London and Queen Mary University of London – created CytoDiffusion around a diffusion-based generative AI classifier. Rather than just learning to separate cell categories, CytoDiffusion models the full range of blood cell morphologies to provide accurate classification with robust anomaly detection.
“Our approach is motivated by the desire to achieve a model with superhuman fidelity, flexibility and metacognitive awareness that can capture the distribution of all possible morphological appearances,” the researchers write.
Authenticity and accuracy
For AI-based analysis to be adopted in the clinic, it’s essential that users trust a model’s learned representations. To assess whether CytoDiffusion could effectively capture the distribution of blood cell images, the team used it to generate synthetic blood cell images. Analysis by experienced haematologists revealed that these synthetic images were near-indistinguishable from genuine images, showing that CytoDiffusion genuinely learns the morphological distribution of blood cells rather than using artefactual shortcuts.
The researchers used multiple datasets to develop and evaluate their diffusion classifier, including CytoData, a custom dataset containing more than half a million anonymized cell images from almost 3000 blood smear slides. In standard classification tasks across these datasets, CytoDiffusion achieved state-of-the-art performance, matching or exceeding the capabilities of traditional discriminative models.
Effective diagnosis from blood smear samples also requires the ability to detect rare or previously unseen cell types. The researchers evaluated CytoDiffusion’s ability to detect blast cells (immature blood cells) in the test datasets. Blast cells are associated with blood malignancies such as leukaemia, and high detection sensitivity is essential to minimize false negatives.
In one dataset, CytoDiffusion detected blast cells with sensitivity and specificity of 0.905 and 0.962, respectively. In contrast, a discriminative model exhibited a poor sensitivity of 0.281. In datasets with erythroblasts as the abnormal cells, CytoDiffusion again outperformed the discriminative model, demonstrating that it can detect abnormal cell types not present in its training data, with the high sensitivity required for clinical applications.
Robust model
It’s important that a classification model is robust to different imaging conditions and can function with sparse training data, as commonly found in clinical applications. When trained and tested on diverse image datasets (different hospitals, microscopes and staining procedures), CytoDiffusion achieved state-of-the-art accuracy in all cases. Likewise, after training on limited subsets of 10, 20 and 50 images per class, CytoDiffusion consistently outperformed discriminative models, particularly in the most data-scarce conditions.
Another essential feature of clinical classification tasks, whether performed by a human or an algorithm, is knowing the uncertainty in the final decision. The researchers developed a framework for evaluating uncertainty and showed that CytoDiffusion produced superior uncertainty estimates to human experts. With uncertainty quantified, cases with high certainty could be processed automatically, with uncertain cases flagged for human review.
“When we tested its accuracy, the system was slightly better than humans,” says first author Simon Deltadahl from the University of Cambridge in a press statement. “But where it really stood out was in knowing when it was uncertain. Our model would never say it was certain and then be wrong, but that is something that humans sometimes do.”
Finally, the team demonstrated CytoDiffusion’s ability to create heat maps highlighting regions that would need to change for an image to be reclassified. This feature provides insight into the model’s decision-making process and shows that it understands subtle differences between similar cell types. Such transparency is essential for clinical deployment of AI, making models more trustworthy as practitioners can verify that classifications are based on legitimate morphological features.
“The true value of healthcare AI lies not in approximating human expertise at lower cost, but in enabling greater diagnostic, prognostic and prescriptive power than either experts or simple statistical models can achieve,” adds co-senior author Parashkev Nachev from University College London.
