Energy harvested from natural or ambient sources is often accompanied by entropy, which the news release describes as randomness. That randomness can make it difficult to convert irregular energy into the predictable electrical power required by batteries, sensors and electronic devices. The researchers set out to examine a protocol intended to overcome this obstacle. Their proposed starting point was to show that deterministic energy harvesting, or DEH, might be impossible, but their analysis instead produced what the team described as a viable quantum solution. Professor Oscar Dahlsten, Associate Professor in CityUHK’s Department of Physics and leader of the research team, said the change in direction—from attempting to prove a limitation to uncovering a possible solution—was among the project’s most surprising developments.

The key feature is that a mixed quantum state can have more than one valid decomposition. In the proposed protocol, a quantum harvester can be charged through multiple pathways corresponding to different intermediate superposition states. If the energy source induces any one of those state transitions, the harvester reaches the same charged final state with certainty. This is why the protocol is described as deterministic: the route taken through the intermediate states may differ, but the final outcome does not. The source can therefore trigger a predictable transfer of energy without a corresponding transfer of entropy. The release presents this as a possible mechanism through which different source states could deliver energy in a controlled manner while avoiding the associated randomness.

The framework offers a theoretical foundation for converting ambient energy into usable power with minimal loss. The researchers suggest that it may eventually inform technologies with lower losses than conventional alternating-current-to-direct-current rectifiers. Potential application areas include Internet-of-Things devices, wearable technologies, smart-city sensors and 6G devices that might draw power from ambient sources. These possibilities remain prospective rather than demonstrated: the supplied evidence concerns the analysis and formulation of a quantum transfer protocol, not a laboratory prototype, field trial or human-facing energy system. The work also indicates that the protocol’s relevance to practical harvesting would require further research and experimental development. The study was supported by a Research Grants Council General Research Fund.

Subject of Research: A theoretical quantum protocol for deterministic energy transfer without entropy

Article Title: CityUHK researchers shatter physical limitations and reveal a quantum pathway to entropy-free energy transfer

Article References: CityUHK researchers shatter physical limitations and reveal a quantum pathway to entropy-free energy transfer. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: quantum energy transfer, mixed quantum states, entropy, deterministic energy harvesting, quantum physics, ambient energy, energy conversion, low-power devices, Physical Review Letters, CityUHK, researchers, shatter

Researchers Yanping Nie and Xianren You, publishing in the open-access journal Results in Engineering, took molybdenum and added small amounts of a high-entropy ceramic, a carbide in which four metals, hafnium, tantalum, niobium, and titanium, share the scene in equal proportions, written chemically as (Hf0.25Ta0.25Nb0.25Ti0.25)C. With just 3 percent by weight of this multi-element ceramic mixed in, the wear rate of the molybdenum composite dropped by 52.8 percent under a 30-newton load and by 28 percent under 40 newtons, while surface hardness climbed by nearly 20 percent. The result points toward a new generation of refractory metal alloys designed for punishing high-temperature wear conditions.

High-entropy ceramics are a relatively young class of materials. Unlike conventional ceramics built around one principal compound, they contain multiple principal elements on the same crystal lattice, generating high configurational entropy. That compositional chaos confers remarkable properties: exceptional hardness, thermal stability, oxidation resistance, and outstanding wear resistance. Engineers have already begun slipping these ceramics into other metallic systems. Reinforcing a tungsten alloy with (HfNbTaZr)C particles, for example, yielded substantial gains in hardness and thermal shock resistance. But until now, incorporating high-entropy ceramics into molybdenum and its alloys remained largely unexplored territory, a gap the new study set out to close.

The synthesis itself reads like a recipe from advanced materials chemistry. The researchers dissolved four metal chlorides, HfCl4, TaCl5, NbCl5, and TiCl4, in a precise 1:1:1:1 molar ratio in absolute ethanol, using triethylamine to neutralize the acid generated during hydrolysis and acetylacetone to chelate and stabilize the metal complexes. A phenolic resin served as the carbon source. After aging, drying, and curing the gel-like precursor, the powder underwent high-temperature carbothermal reduction at 1600 degrees Celsius under argon, producing a single-phase high-entropy carbide. The ceramic powder was then blended with commercial molybdenum powder by high-energy ball milling and consolidated by spark plasma sintering, ramped to 2000 degrees Celsius at 100 degrees per minute under 50 megapascals of pressure.

Microscopy confirmed that the strategy worked at the finest scales. Scanning electron microscopy with elemental mapping showed the ceramic particles uniformly embedded in the molybdenum matrix without micro-voids, cracks, or large-scale agglomeration. Crucially, the constituent elements of the ceramic stayed localized within the particles rather than diffusing wholesale into the matrix, confirming chemical stability during the brutal sintering process. Transmission electron microscopy went further, revealing a distinct interfacial transition layer between the ceramic and the metal, roughly continuous and crack-free, with lattice fringes matching the (111) plane of the high-entropy carbide and the (110) plane of body-centered cubic molybdenum. The gradual diffusion gradient across that layer showed it was not a mechanical gap but a chemically continuous zone forged by atomic interdiffusion, the microscopic secret behind robust load transfer.

Electron backscatter diffraction added another layer of insight. The KAM values, a proxy for dislocation density, rose systematically with ceramic addition, from 0.92 degrees in pure molybdenum to 0.97 and 1.01 degrees in the composites, consistent with the Taylor hardening mechanism in which more dislocations mean greater hardness. The pure molybdenum sample was dominated by low-angle grain boundaries, which make up 75.8 percent of the total and do little to block dislocation motion. Doping with the ceramic rebalanced that boundary structure, particularly in the 3 percent composite, facilitating more homogeneous deformation and mitigating localized stress. Interestingly, the ceramic phase also coarsened the grains, pinning boundaries and reducing nucleation during sintering, yet the fine dispersion of hard particles more than compensated.

The hardness measurements told a clear story. Pure molybdenum reached about 212 HV, while the 3 percent and 5 percent composites hit 253.3 HV and 264.9 HV, increases of roughly 19.4 percent and 24.9 percent. But hardness alone does not guarantee wear resistance, and the friction tests under a silicon nitride counterbody revealed a subtler picture. At 30 newtons, pure molybdenum actually had the lowest friction coefficient, about 0.37, because its low shear resistance lets it slide easily. The 3 percent composite registered 0.45 and the 5 percent sample 0.55. Yet the 3 percent composite lost the least material, its wear rate falling to 0.75 × 10⁻⁷ mm³/N·m, nearly half that of pure molybdenum. The hard, well-bonded particles resist plastic deformation and shield the underlying matrix, even as they plow against the counterface and raise friction.

The 5 percent composite told the opposite story. Its high friction coefficient came hand in hand with a high wear rate, evidence that excessive ceramic loading weakens interfacial bonding and promotes particle detachment. Those liberated hard particles then circulate between the surfaces as three-body abrasives, grinding away material from both sides. Optical microscopy of the counterpart balls confirmed the damage asymmetry: at 30 newtons, the ball sliding against the 3 percent composite showed the smallest wear scar diameter, 329 micrometers versus 381 micrometers against pure molybdenum, meaning the composite protected the opposing surface too. Three-dimensional profilometry and cross-sectional depth measurements reinforced the trend, with the 3 percent composite showing the shallowest wear scars, about 4.96 micrometers deep at 30 newtons compared with 8.3 micrometers for pure molybdenum.

Load changed everything. At 50 newtons, the trend reversed dramatically. The wear rates of both composites spiked, reaching 2.13 × 10⁻⁷ and 1.94 × 10⁻⁷ mm³/N·m, exceeding even that of pure molybdenum. Scanning electron microscopy of the worn surfaces revealed why: the excessive normal stress extruded and fractured the ceramic particles out of the matrix, and the detached fragments became aggressive three-body abrasives that carved deep pits and irregular grooves into the surfaces. The wear mechanism shifted from adhesive and oxidative wear at low loads, where transferred silicon dioxide from the counterbody even formed a partially protective scale, to severe abrasive wear at high loads. The protective effect of the ceramic dispersion, so powerful at moderate loads, simply could not survive the particle pull-out triggered by excessive stress.

The study’s takeaway is a lesson in balance. Moderate addition of a high-entropy carbide, around 3 percent by weight, delivers a synergistic combination of fine particle dispersion, robust interfacial integrity, elevated dislocation density, and a balanced grain boundary structure, suppressing plastic deformation and crack initiation while keeping friction manageable. Push the ceramic content higher, and weakly bonded particles turn from shields into saboteurs. For engineers designing molybdenum alloys for demanding moderate-load, high-temperature wear applications, the message is precise: the optimal reinforcement is not the maximum reinforcement. The researchers note that future work will focus on cross-sectional characterization of the wear scars to further illuminate the underlying mechanisms, but the core finding already stands as a compelling demonstration that high-entropy ceramics can transform the tribological performance of refractory metals.

Subject of Research: Enhancement of wear resistance in Mo-based composites through moderate addition of (Hf 0.25 Ta 0.25 Nb 0.25 Ti 0.25 )C high entropy ceramics

Article Title: Enhancement of wear resistance in Mo-based composites through moderate addition of (Hf 0.25 Ta 0.25 Nb 0.25 Ti 0.25 )C high entropy ceramics

Article References: Nie, Y., & You, X. (2026). Enhancement of wear resistance in Mo-based composites through moderate addition of (Hf0.25Ta0.25Nb0.25Ti0.25)C high entropy ceramics. Results in Engineering, 32, Article 113570. https://doi.org/10.1016/j.rineng.2026.113570

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113570

Keywords: Enhancement, wear, resistance, Mo-based, composites, moderate, addition, high, entropy, ceramics, scientific research

The threshold approach rests on a deceptively simple piece of decision theory. For any suspected diagnosis, a clinician can compare the expected benefit of treatment against the expected harm. If the benefit-to-harm ratio is high, treatment is justified even at low probability of disease; if the ratio is low, treatment is warranted only at high probability. These two probabilities define the testing threshold and the treatment threshold. Below the testing threshold, the diagnosis is too improbable to pursue, and further testing adds more risk than information. Above the treatment threshold, the diagnosis is probable enough that the benefits of therapy outweigh the harms of being wrong. Between the two thresholds lies the zone of diagnostic uncertainty, where additional tests are genuinely useful because their results can move the probability across a boundary that changes management.

Rohlfsen and Parsons contend that this elegant structure is being eroded not by new evidence but by the sheer volume of modern clinical information. They invoke entropy in its information-theoretic sense, the average uncertainty contained in a probability distribution. Every laboratory result, imaging study, and vital sign carries information, but the diagnostic value of any single result depends on what the clinician already believes. As the number of available data points multiplies, the average uncertainty in the aggregate picture can rise even when each individual result seems informative. The authors suggest that clinicians and educators have begun to mistake the accumulation of data for the reduction of uncertainty, a confusion that the threshold approach was designed to prevent.

The letter draws on a substantial body of empirical work on diagnostic reasoning. Peter Han, William Klein, and Nikhil Arora, writing in Medical Decision Making in 2011, developed a conceptual taxonomy of uncertainty in health care, distinguishing between the uncertainty of probability, the ambiguity of evidence, and the complexity of the clinical situation. Their framework makes clear that not all uncertainty is statistical. A clinician may know the sensitivity and specificity of a test with precision yet remain uncertain about how those parameters apply to a particular patient whose presentation does not match the populations in which the test was validated. The threshold approach addresses only the probabilistic component, and Rohlfsen and Parsons argue that medical education has overemphasized that component at the expense of the others.

Empirical studies have documented the consequences. Andrew Meyer, Vishal Payne, David Meeks, Rakesh Rao, and Hardeep Singh, in a 2013 vignette study published in JAMA Internal Medicine, examined physicians’ diagnostic accuracy, confidence, and resource requests. They found that diagnostic confidence was often poorly calibrated with respect to actual accuracy, and that physicians frequently requested additional resources even when their stated confidence in a diagnosis was high. The finding is difficult to reconcile with a purely rational threshold model, in which resource requests should scale with the distance between the current probability and the relevant threshold. The authors of the letter suggest that the discrepancy reflects a failure to teach clinicians where their thresholds actually lie, leaving them to set boundaries by intuition rather than by explicit reasoning.

Jeffrey Stojan, Mahan Daniel, Sara Hartley, and Larry Gruppen, writing in Medical Education in 2022, developed a threshold model of dealing with uncertainty in clinical reasoning, examining the roles of experience and task framing. Their work suggests that experienced clinicians do not simply have better-calibrated thresholds; they also frame diagnostic tasks differently, sometimes treating a case as a learning exercise and sometimes as a high-stakes decision, with corresponding shifts in how much uncertainty they are willing to tolerate before acting. The letter argues that this variability is not a defect but a feature of expert judgment, and that pedagogy should make it explicit rather than pretend that a single threshold applies to every situation.

Rohlfsen and Parsons also reference the 5L framework of diagnostic reasoning, developed by Olivia Brumfield, Ahmed Mohamed, Sheng He, and Cory Rohlfsen and published in the Journal of Hospital Medicine in 2026. The framework provides a stepwise scaffold to support clinician educators, breaking diagnostic reasoning into discrete stages that can be taught and assessed. The letter situates the threshold approach within this pedagogical structure, suggesting that the 5L framework gives educators a language for discussing where a diagnostic probability stands relative to a decision threshold, and what kind of evidence would move it. Without such a language, the authors contend, discussions of diagnostic reasoning tend to collapse into vague appeals to clinical gestalt, which students cannot replicate and which resist assessment.

The letter also invokes the ADEPT study protocol, a multicenter prospective quality and safety program described by Jeffrey Schnipper and colleagues in the Journal of Hospital Medicine in 2023, designed to improve diagnostic processes in medical inpatients through prevention and teamwork. The letter suggests that programs like ADEPT represent the systems-level counterpart to the threshold approach, embedding diagnostic reasoning into the structure of hospital care rather than leaving it to the individual clinician. The authors argue that diagnostic excellence requires both the individual judgment of the threshold approach and the organizational support of programs that make diagnostic processes visible, auditable, and improvable.

The implications extend beyond internal medicine. Rohlfsen and Parsons write in the context of predictive medicine and the philosophy of medicine, subjects that the journal’s machine-learning-driven related-articles system associates with the letter. The authors suggest that as diagnostic algorithms become more prevalent, the threshold approach may need to be renegotiated. An algorithm can output a probability, but it cannot set a threshold, because threshold setting depends on the benefit-to-harm ratio of treatment, which is a value judgment about what matters to a particular patient. The letter implies that clinicians who delegate probability estimation to algorithms must retain the judgment required to set thresholds, and that medical education should teach that judgment explicitly rather than assume it emerges from exposure to clinical work.

The letter, published on 8 September 2026, received no funding sources and reports no conflicts of interest. Rohlfsen wrote the initial draft and Parsons reviewed and edited it, with both contributing to the conceptualization of the submission. The work is situated within a tradition of clinical decision theory that stretches back to the Pauker and Kassirer paper, and the authors suggest that the tradition remains vital precisely because the problems it addresses have not been solved by technology. The entropy of modern clinical data has increased, the authors argue, but the fundamental question of when a diagnosis is probable enough to act upon remains a matter of judgment, and judgment is something that must be taught.

Subject of Research: Beyond Entropy: Decision Threshold, Judgment, and Pedagogy

Article Title: Beyond Entropy: Decision Threshold, Judgment, and Pedagogy

Article References: Rohlfsen, C., & Parsons, A. S. (2026). Beyond Entropy: Decision Threshold, Judgment, and Pedagogy. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10753-4

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10753-4

Keywords: Beyond, Entropy, Decision, Threshold, Judgment, Pedagogy, scientific research

Tuchinda’s winning project, titled The Role of Entropy in the Interfacial Cohesion of Metals and Alloys, sets its sights on a factor that conventional models of material failure have long treated as an afterthought: entropy. Entropy is a thermodynamic property related to the number of possible arrangements and motions that atoms can adopt within a material at finite temperatures. That may sound abstract, but it describes the actual environment that engineering materials face in real processing routes and real applications, where heat, vibration and atomic shuffling are unavoidable facts of life. Although scientists have spent decades studying how chemical elements accumulate along crystal boundaries and alter their strength, the entropic contributions are frequently excluded from predictive calculations altogether, simply because they are difficult and computationally expensive to account for. Tuchinda’s project aims to close that gap by developing a more complete understanding of how atomic arrangements, chemical composition and temperature jointly determine the strength of the interfaces that hold metals together.

The practical stakes of this question are enormous. Metals and alloys contain microscopic crystals separated by internal boundaries, and these boundaries can become vulnerable points where fractures begin. Chemical elements that segregate along these boundaries can either strengthen the connections between crystals or weaken them dramatically, influencing whether a material withstands mechanical stress or becomes susceptible to sudden brittle failure. Sulfur embrittling steel at high temperature, or beneficial elements toughening a superalloy, are two sides of the same atomic-scale coin. Understanding this type of mechanism, Tuchinda explained, will help materials scientists and engineers design both new processing routes and alloys that are resistant to fracture, such as advanced steels and refractory alloys intended for nuclear applications. In other words, the research could contribute to the development of stronger, more reliable materials for the demanding energy environments where failure is simply not an option.

To attack the problem, Tuchinda will deploy a toolkit that reflects the modern transformation of materials science. His approach combines advanced computer simulations, thermodynamic modeling and machine-learning-assisted techniques to disentangle the interplay between atomic structure, chemistry and temperature at internal interfaces. At NJIT he leads the Imperfection Engineering Group, a research team that uses accelerated computer modeling and simulations to investigate the structure of imperfections in crystalline materials. The group studies grain boundaries, triple junctions, which are the locations where three crystals meet, and other structural imperfections within crystalline materials, blending computational materials science, thermodynamics and data-driven modeling to understand how these features affect material behavior. Rather than treating defects as nuisances to be eliminated, the group treats them as designable features whose properties can be engineered.

The DOE award builds on a substantial foundation of published research. In a 2026 review published in the journal Acta Materialia, Tuchinda and collaborators examined advances in computational models that predict how chemical elements accumulate along grain boundaries. Their review highlights how accounting for differences in atomic environments and thermodynamic properties, including entropy, can improve predictions of material behavior and inform the design of more stable alloys. The message of that review is that the field’s predictive models have been leaving measurable accuracy on the table by ignoring entropic effects, and that correcting this omission could change how alloy designers think about stability at elevated temperatures.

Tuchinda’s earlier research, carried out during his doctoral studies at the Massachusetts Institute of Technology, investigated how alloy composition and atomic-scale interactions influence material strength and stability. Those studies examined grain-boundary embrittlement in aluminum alloys, developed large computational databases to predict how chemical elements affect interfacial cohesion, and advanced machine-learning-assisted methods for understanding atomic interactions across different alloy systems. Each of these strands feeds directly into the new DOE project. The databases provide the raw predictive machinery, the machine-learning methods accelerate the exploration of vast chemical spaces, and the embrittlement studies supply the physical intuition about how segregation translates into fracture.

The new project will also expand the scope of the research well beyond grain boundaries. Engineering alloys contain many other types of internal interfaces, including boundaries between different material phases, interfaces involving intermetallic compounds, and surfaces where reinforcing particles meet the surrounding metal matrix. The research will examine how atomic arrangements and their stability at different temperatures influence the accumulation of chemical elements at these interfaces and the strength of the connections between them. This is territory where scientists still lack comprehensive models to explain and predict what happens, which means the project could establish explanatory frameworks that do not currently exist. Success would give alloy designers a unified picture of interfacial cohesion across the full menagerie of interfaces found in real engineering materials, not just the classical grain boundaries that have dominated the literature.

The award carries institutional as well as scientific weight. This DOE project will help establish my lab and support graduate students who will further develop the future STEM workforce, and help mentor undergraduate students at NJIT, Tuchinda said. The funding arrives at a formative moment for a young research group, providing the sustained five-year horizon that fundamental computational research requires. The DOE Early Career Research Program, administered through the Office of Science, supports scientists at U.S. universities, national laboratories and user facilities pursuing fundamental research in areas central to the department’s mission, and the 2026 cohort draws its 67 scientists from 39 universities and 11 DOE national laboratories and user facilities. Tuchinda’s project was selected through the DOE Office of Basic Energy Sciences, which supports research into the fundamental properties and behavior of matter.

Tuchinda’s trajectory traces a clear line from Bangkok to Newark. He earned his bachelor’s degree in metallurgical and materials engineering from Chulalongkorn University in Thailand in 2018, then moved to MIT, where he completed his Ph.D. in materials science and engineering in 2023. Five years later he stands at the head of his own laboratory, federally funded to pursue the entropic physics of the seams that hold metals together. If his project succeeds, the atomic-scale interactions that determine whether a nuclear alloy toughens or shatters will become predictable quantities in the equations of alloy design, transforming a long-overlooked thermodynamic property into a practical engineering tool for the materials that power the nation’s energy future.

Subject of Research: Entropy-driven interfacial cohesion and failure prediction in metals and alloys

Article Title: NJIT professor selected for DOE early career award to advance research on metal failure

Article References: NJIT professor selected for DOE early career award to advance research on metal failure. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: materials science, metallurgy, entropy, grain boundaries, DOE Early Career Research Program, NJIT, alloy design, brittle fracture, computational materials science, machine learning, professor, selected

Darwin himself wrestled with the question. In an 1838 notebook entry, he wrote that as forms became complicated they opened fresh means of adding to their complexity, but that there was no necessary tendency in simple animals to become complicated. Edelman takes this nuance as his starting point: complexity increase is a tendency, not a universal law, and it need not apply across the board. Bacteria and archaea have coexisted with metazoans for hundreds of millions of years, and complexity has risen and fallen in individual lineages.

Edelman identifies at least three families of mechanisms that can drive complexity upward. The first is entirely passive: in any evolutionary system with variation and heredity, complexity tends to increase simply because it is bounded from below. A random walk in complexity space, starting near the minimum, can only drift upward in the long run. This insight underlies the Zero-Force Evolutionary Law proposed by McShea and Brandon, which holds that rising diversity and complexity are the null expectation in the total absence of selection.

The second mechanism is entropic. Complexity growth can proceed concomitantly with increasing entropy, the probable direction of a system’s progression through its configuration space. Edelman focuses on Boltzmann entropy, where the entropy of a macrostate is proportional to the logarithm of the number of microstates it contains. Transitions from macrostates with fewer microstates to those with more are more probable, so the succession of macrostates over time tends toward higher entropy. Crucially, this probabilistic dynamics does not require the system to be closed or to approach equilibrium.

The apparent paradox of ordered biological structures arising under a disordering principle dissolves once the word order is confined to its proper realm. As Wicken argued in 1979, biological structures are not ordered in the information-theoretic sense; they are organized. The Second Law acts to dissipate order, not organization, and organization tends to accumulate at the expense of the former. Brooks and Wiley formalized this idea, noting that when the phase space of possibilities expands faster than the system can fill it, increasing entropy can be accompanied by the emergence of organized structure.

The third family of mechanisms involves selection. Saunders and Ho showed in 1976 that at a local fitness peak, random removal of a component must make the system less fit, while random addition allows the possibility of an increase. This asymmetry acts as a ratchet, amplifying passive complexity increases. Adami and colleagues demonstrated in simulated evolution that selection forces genomes to behave as a natural Maxwell Demon, within a fixed environment, genomic complexity is forced to increase. Selection can also open new possibilities for adaptation, expanding the effective genetic phase space and echoing Darwin’s observation that complexity breeds complexity.

Edelman then extends these ideas to consciousness, focusing on basic phenomenal awareness, the simplest variety of consciousness, exemplified by simple perceptions such as the taste of a wine or the smell of a flower. According to his Dynamical Emergence Theory, this level of awareness consists in dynamically emergent diagnosis by a cognitive system of its own representational states, what he calls autodiagnosis. The stream of awareness is identified with the succession of emergent metastable macrostates in the system’s state space, and these macrostates are non-arbitrary because they are equivalence classes of future predictability.

Basic phenomenal awareness can become more complex over evolutionary time or through learning. Edelman illustrates the evolutionary scale with the progression from simple photosensitivity in planarian eyespots to the imaging camera eye of frogs, cortical shape perception in mammals, and color perception. Learning-induced complexification is exemplified by wine tasting expertise, where a taster progresses from distinguishing red from white wine to identifying a specific variety, vineyard, and vintage. On an abstract computational level, both types of complexification increase the topological complexity of the system’s trajectory space, and the acquisition of exemplars and categories makes the process entropic, as the system’s total entropy increases with learning.

As a test case, Edelman proposes reservoir computing, an architecture consisting of a reservoir of coupled neurons with nonlinear dynamics and a linear readout circuit. This architecture naturally maps onto the olfactory circuit of insects such as the locust and fruit fly, where recurrently interconnected projection neurons in the antennal lobe project to Kenyon cells in the mushroom body. Edelman outlines several experiments using this model to test whether the entropy and topological complexity of the system’s dynamics tend to increase over evolutionary time and through learning.

The review concludes that if one seeks a general tendency rather than a universal law, complexity increase emerges as a non-trivial principle of evolution, underwritten by several distinct mechanisms. This principle applies not just to the evolution of species morphologies but also to the evolution and learning of their capacity for phenomenal discernment, which constitutes the most basic level of consciousness. Edelman notes that the growth of complexity is expected to accelerate when greater complexity facilitates evolvability, and that biology co-evolving with culture provides a cumulative, self-reinforcing catalyst. The scope of such an approach could extend even to cosmological selection hypotheses, though that extension awaits future research.

Subject of Research: On the evolutionary dynamics of complexity and consciousness

Article Title: On the evolutionary dynamics of complexity and consciousness

Article References: Edelman, S. (2026). On the evolutionary dynamics of complexity and consciousness. Frontiers in Complex Systems, 3, Article 1672525. https://doi.org/10.3389/fcpxs.2025.1672525

Image Credits: AI Generated

DOI: 10.3389/fcpxs.2025.1672525

Keywords: evolution, complexity, consciousness, entropy, phenomenal awareness, Zero-Force Evolutionary Law, Dynamical Emergence Theory, reservoir computing, natural selection, thermodynamics, Shimon Edelman, Frontiers in Complex Systems

The core innovation of the framework lies in its refusal to rely on any single performance statistic. Conventional assessments often hinge on one or two measures of fit, such as root mean square error or correlation, which can hide crucial failures in how a model reproduces the character of a climate variable. The entropy-based composite index instead combines five complementary indicators that capture different dimensions of model behavior: distributional similarity, temporal complexity, spatial structure, seasonal behavior, and temporal persistence. Distributional similarity asks whether a model reproduces the overall statistical distribution of observed values, for example whether extreme rainfalls and long dry spells occur with realistic frequency. Temporal complexity examines the unpredictable, fluctuating quality of the observed time series, borrowing concepts such as approximate entropy and sample entropy that were originally developed for analyzing physiological signals like heart rate variability. Spatial structure measures whether the model gets the geographic pattern right, and seasonal behavior checks whether the annual cycle of monsoon and dry seasons is faithfully simulated. Temporal persistence, finally, evaluates how well the model captures the tendency of climate variables to carry memory from one period to the next.

Entropy serves as the mathematical backbone of the entire scheme. In information theory, entropy quantifies the uncertainty or information content of a system, a concept introduced by Claude Shannon in his landmark 1948 work on communication. When applied to model evaluation, entropy-based weighting offers a way to determine the relative importance of each indicator without human judgment entering the calculation. Indicators that discriminate strongly between models, that is, those on which the models disagree most, receive higher weights, while indicators on which all models perform similarly contribute less to the final ranking. This contrasts sharply with traditional multicriterion decision-making approaches in which weights are assigned subjectively by experts, a practice that has been shown in earlier studies to influence which models rise to the top. By letting the data itself dictate the weights, the Kharagpur team removes a persistent source of bias from the model selection process.

The study put ten CMIP6 models through this evaluative gauntlet across the entire Indian subcontinent, comparing their simulations against observational data at a one-degree grid scale for three variables: precipitation, maximum temperature, and minimum temperature. The observational datasets came from the India Meteorological Department, providing a ground truth rooted in the subcontinent’s remarkably diverse climate. The choice of grid-scale evaluation matters because averaging performance over the whole country would mask the spatial heterogeneity that dominates Indian climate. A model that excels over the humid Ganges basin might flounder over the Deccan plateau, and the framework was specifically designed to expose such regional contrasts rather than bury them in a national average.

The results reveal a striking asymmetry between how models handle water and how they handle heat. For precipitation, the rankings showed pronounced spatial fragmentation, with no single model emerging as a universal champion. Instead, NorESM2-MM, MIROC6, and CNRM-CM6-1 each performed better in different agro-climatic zones, reflecting the notorious difficulty of simulating monsoon dynamics, orographic rainfall, and convective processes. The practical consequence is significant: in several regions, particularly over the Himalayas and the arid zones, the authors conclude that ensemble-based selection is required, meaning that no individual model can be relied upon alone and that a carefully weighted combination of models offers the most defensible basis for projections. This finding resonates with a broader literature on the challenges of combining climate model projections, which has long recognized that precipitation remains among the hardest variables for global models to pin down.

Temperature variables, by contrast, displayed much clearer patterns of dominance. For maximum temperature, the models ACCESS-CM2 and CNRM-CM6-1 performed consistently well across the estimation tasks, suggesting that their representations of surface energy balance and diurnal heating are more robust over the subcontinent. For minimum temperature, MRI-ESM2-0 stood out as the best performer across multiple regions, a notable result given that minimum temperatures govern many impacts of practical concern, from heat stress at night to frost risk in agricultural areas. The contrast between the fragmented precipitation picture and the coherent temperature rankings offers actionable guidance: researchers projecting future heat extremes in India can identify a single best-performing model with confidence, while those working on rainfall must think regionally and in terms of ensembles.

One of the most technically interesting results concerns which of the five indicators actually mattered most. The entropy-derived weights highlighted the dominant role of temporal complexity and persistence indicators, especially for temperature variables. This suggests that what separates a good climate model from a mediocre one over India is not merely whether it gets the average right, but whether it reproduces the intricate day-to-day and season-to-season texture of the climate record, including its memory and its irregularity. Models that smooth away the natural variability of the observed record may look acceptable under simple error metrics while failing to capture the dynamics that matter for extremes, and the entropy weighting scheme automatically penalizes such behavior by assigning those indicators greater influence.

The implications reach well beyond the Indian subcontinent. The framework’s design, being objective, reproducible, and scalable to any region with adequate observational data, makes it a candidate for adoption in other countries grappling with the same model selection dilemma. Climate impact and adaptation studies increasingly depend on downscaling and bias correction of global model output, and an erroneous initial choice of model can propagate misleading projections into hydrological models, crop simulators, and infrastructure planning tools. Previous efforts to rank CMIP6 models for India using techniques such as TOPSIS and other multicriterion analyses have contributed valuable insights, but the entropy-based composite index adds a layer of methodological rigor by making the weighting process itself data-driven and by integrating a broader suite of statistical diagnostics into a single transparent score.

For the working researcher, the study functions as something like a consumer report for climate models, delivering zone-specific recommendations rather than a one-size-fits-all verdict. A hydrologist assessing flood risk in a Himalayan basin, for instance, would consult the framework’s findings for that specific agro-climatic zone and find guidance on whether a single model such as NorESM2-MM or an ensemble is the appropriate basis for projections, and which models should form that ensemble. The authors emphasize that the framework provides robust guidance for climate impact and adaptation studies over India and elsewhere, and the emphasis on agro-climatic zones aligns the technical output with the administrative and agricultural units through which adaptation decisions are actually made. As CMIP6-based projections continue to feed into national climate assessments and international reports, tools that make model selection defensible, transparent, and regionally honest are likely to become as essential as the models themselves. The entropy-based composite index represents a meaningful step toward treating model selection with the same statistical care that climate science applies to every other stage of its projections.

Subject of Research: Objective ranking of CMIP6 General Circulation Models using an entropy-based composite index over India

Article Title: A novel entropy-based composite index for ranking General Circulation Models under CMIP6

Article References: Rawat, K., & Sahoo, B. (2026). A novel entropy-based composite index for ranking General Circulation Models under CMIP6. Theoretical and Applied Climatology, 157(11), Article 717. https://doi.org/10.1007/s00704-026-06642-9

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06642-9

Keywords: CMIP6, General Circulation Models, entropy, composite index, climate model evaluation, Indian monsoon, precipitation, temperature, agro-climatic zones, model weighting, Theoretical and Applied Climatology, IIT Kharagpur

The central problem the work addresses is one of representation rather than mechanism. Modern neuroscience can measure neural activity with extraordinary precision, yet most quantitative analyses treat each observable — signal amplitude, spectral power, entropy, phase relationships, cross-frequency coupling, and recurrence structure — as an independent descriptor. No single measure captures the full evolution of large-scale brain dynamics. Klebba’s proposal is that these established observables can instead be interpreted as complementary properties of a single evolving neural state, represented mathematically by a complex-valued function in which instantaneous amplitude and phase unfold continuously over time.

The mathematical foundation is deliberately classical. Using the Hilbert transform, an oscillatory EEG signal can be converted into an analytic signal — a complex-valued representation whose magnitude gives instantaneous amplitude and whose argument gives instantaneous phase. This is a standard tool in electrophysiology, and the study is emphatic that nothing quantum is being claimed. The inspiration from Everett is strictly structural: his formalism showed how multiple evolving configurations could be represented within a single state space without requiring collapse into one physical state. The framework adopts that representational strategy, describing the neural state as a superposition of multiple simultaneously active neural configurations, each a transient pattern of coordinated activity whose relative contribution changes continuously.

Crucially, these configurations are not proposed as separate conscious entities or parallel minds. They are mathematical constructs describing the many interacting patterns of neural activity that coexist in one biological brain — sensory processing, memory retrieval, attentional control, motor planning, and predictive inference all running in parallel across distributed networks. The framework suggests, as a theoretical hypothesis rather than an empirical result, that a unified thought might correspond to the transient dominance of one configuration within this evolving landscape, while alternatives remain dynamically active outside immediate awareness. Whether that hypothesis holds is explicitly left for future investigation.

To ground the framework empirically, the study analyzed a publicly available EEG dataset collected through a collaboration between the Neurocomputing Laboratory at IIT Delhi and the Neurology & Sleep Center in New Delhi. Recordings from ten epilepsy patients, acquired at 200 Hz with standard international 10–20 electrode placement, were divided into 50 interictal, 50 preictal, and 50 ictal segments — 150 recordings in total, each 5.12 seconds long. The dataset was chosen because it contains clearly defined transitions between distinct physiological states, providing a natural testbed for asking whether multiple electrophysiological observables change in a coordinated manner.

The results were striking in their consistency. Root-mean-square amplitude increased from about 23.5 microvolts during interictal recordings to roughly 139 microvolts during seizures, one of the largest effects in the study with an effect size of ε² = 0.658. Spectral power redistributed dramatically toward lower frequencies: delta-band activity rose from 40 percent of total power to 75.4 percent during seizures, while theta, alpha, and gamma contributions fell sharply. All three entropy measures — spectral, permutation, and sample entropy — declined significantly, indicating progressively reduced signal complexity as seizures took hold, with sample entropy showing one of the largest effects at ε² = 0.586.

Temporal organization changed as well. Alpha-band phase concentration increased significantly during ictal recordings, and theta–gamma phase-amplitude coupling, quantified with the modulation index and validated against surrogate data, differed significantly between interictal and seizure-associated states. Recurrence quantification analysis, performed after reconstructing each signal’s state space through time-delay embedding, revealed that neural trajectories became more persistent and less mobile during seizures: mean diagonal line length grew from 4.05 to 7.70, determinism increased, and mean state-space displacement fell from 0.422 to 0.291. Every major analytical domain showed statistically significant differences across seizure states, assessed with Kruskal–Wallis tests, Holm-corrected post hoc comparisons, effect sizes, and bootstrap confidence intervals.

The author is careful to situate these findings relative to existing theories of consciousness. The results are broadly consistent with Integrated Information Theory’s emphasis on large-scale integration, but the framework does not attempt to compute Φ or propose an alternative measure of consciousness. It likewise differs from electromagnetic field theories such as CEMI and Susan Pockett’s formulation: the neural field here is a mathematical representation of measurable dynamics, not a claim that the brain’s physical electromagnetic field is the substrate of experience. The framework also engages the concept of metastability — the tendency of complex neural systems to occupy transiently organized configurations while retaining the capacity to reorganize — as the theoretical context in which the evolving state representation should be understood.

The study acknowledges its limitations candidly. The recordings were single-channel, so they cannot resolve the distributed cortical interactions that the proposed configurations would represent; the data come from pathological brain states that cannot be assumed to mirror healthy cognition; and neither the individual configurations nor the proposed neural coordination functional were directly estimated from the data. The empirical contribution is the demonstration that established observables change in a coordinated, multidimensional way during seizure progression — a quantitative foundation, not a validation, of the formalism.

What the work ultimately offers is a research program rather than a finished theory. Future investigations using high-density EEG, magnetoencephalography, intracranial recordings, and computational state-space reconstruction could test whether candidate neural configurations can be empirically identified, whether transitions among them correspond to changes in attention or cognition, and whether similar coordinated dynamics appear in sleep, anesthesia, meditation, and disorders of consciousness. The framework even suggests applications beyond neuroscience, potentially informing biologically inspired artificial intelligence systems that rely on dynamical representations of high-dimensional state spaces. Whether wave-function formalism proves genuinely useful for understanding the brain will depend not on its elegance but on whether it generates predictions that competing descriptions cannot — a question the author deliberately frames as open, empirical, and now, at least, formally posed.

Subject of Research: A wave-function-based mathematical framework for representing evolving neural dynamics measured with EEG during epileptic seizure states

Article Title: Wave-function formalism for evolving neural dynamics: a mathematical framework informed by EEG

Article References: Klebba, M. (2026). Wave-function formalism for evolving neural dynamics: a mathematical framework informed by EEG. Frontiers in Computational Neuroscience, 20, Article 1812909. https://doi.org/10.3389/fncom.2026.1812909

Image Credits: AI Generated

DOI: 10.3389/fncom.2026.1812909

Keywords: EEG, neural dynamics, wave-function formalism, epilepsy, computational neuroscience, Hilbert transform, entropy, phase-amplitude coupling, recurrence analysis, metastability, state space, consciousness

The two techniques in question are both built on diffusion-weighted imaging, or DWI, a cornerstone of modern oncological MRI. DWI works by sensitizing the MRI signal to the random Brownian motion of water molecules. In dense, cellular tumors, water molecules bump into cell membranes and extracellular matrix far more often than they do in loose, healthy tissue, so water diffusion is restricted and the tumor lights up brightly on DWI images. From these images radiologists calculate a quantity called the apparent diffusion coefficient, or ADC, which quantifies diffusion numerically. Lower ADC values generally correspond to more cellular, more aggressive tissue, and ADC thresholds are widely used to help distinguish malignant from benign lesions, to assess treatment response, and to characterize tumors across the body.

Conventional DWI, which the researchers call General-DWI, uses a well-established acquisition method known as single-shot echo-planar imaging combined with parallel imaging. It covers the entire field of view in one rapid sweep, which makes it fast and robust, but it comes with an inherent trade-off: the large field of view and the physics of echo-planar readouts introduce geometric distortion, blurring, and image noise, particularly in the abdomen, where breathing motion and nearby air-filled organs degrade image quality. For gastric cancer specifically, where lesions can be small and sit amid a tangle of motion-prone anatomy, those artifacts can blur the very margins clinicians most need to see.

Enter Zoom-DWI, short for zonally oblique multi-slice diffusion-weighted imaging. Instead of exciting the whole imaging volume and then reconstructing the slice of interest, Zoom-DWI uses a targeted, two-dimensional radiofrequency excitation scheme that excites only a thin, obliquely oriented slab containing the anatomy of interest. By never exciting tissue outside that zone in the first place, the technique avoids the aliasing and fold-over artifacts that conventional acquisitions must otherwise suppress with parallel imaging acceleration. The theoretical payoff is a smaller effective field of view, reduced distortion, and crisper anatomical detail, all without requiring the patient to hold their breath, since the sequence is acquired during free breathing. The technique has shown promise in other organs, including prostate and breast imaging, but its behavior in gastric cancer had not been systematically benchmarked against conventional DWI until now.

The research team, led by radiologists at Daping Hospital of Army Medical University in Chongqing, conducted a single-center retrospective study of 70 patients with histopathologically confirmed gastric cancer who underwent abdominal MRI between March 2023 and October 2024. Each patient received both sequences: the conventional free-breathing, fat-suppressed General-DWI and the targeted Zoom-DWI. The investigators then evaluated the two techniques on three fronts. First, they scored image quality qualitatively on a five-point scale, with two independent readers assessing features such as tumor boundary delineation and image noise. Second, they measured mean ADC values within carefully drawn tumor regions of interest. Third, they computed entropy metrics, a texture-analysis measure that captures the heterogeneity, or spatial irregularity, of the signal within the tumor, which some researchers hope can serve as an imaging proxy for tumor aggressiveness.

The headline finding on image quality was a genuine split decision. Zoom-DWI scored significantly higher for tumor boundary delineation, with a mean score of 3.78 compared with lower scores for the conventional technique, a difference that was highly statistically significant. In plain terms, radiologists could see where the tumor ended and healthy tissue began more clearly with the zoomed approach, which matters enormously for staging, treatment planning, and targeting biopsies. But the sharper view came at a cost: Zoom-DWI performed significantly worse on image noise, scoring 3.76 on the noise assessment, indicating that the targeted excitation approach produces grainier, less clean images than the conventional method. This trade-off between spatial clarity and noise is a familiar refrain in MRI physics, but quantifying it in gastric cancer provides practical guidance for radiologists deciding which sequence to deploy.

The quantitative results revealed something more consequential. General-DWI produced significantly higher ADC values than Zoom-DWI: 1.43 versus 1.33 in the standard units of ten-thousandths of a square millimeter per second, a difference that was highly significant. To dig deeper, the team applied linear mixed-effects modeling, a statistical framework that can separate true sequence effects from patient-to-patient variability. The modeling revealed a systematic ADC underestimation bias of 0.096 associated with Zoom-DWI, meaning the zoomed technique consistently reads diffusion as slightly more restricted than the conventional method does in the same tumor. This is not random scatter; it is a predictable, directional shift baked into the physics of the acquisition.

Why does this matter so much? Because ADC values are not just descriptive numbers, they are diagnostic instruments. Radiologists and researchers have accumulated years of threshold-based criteria, for example cutoffs used to separate malignant from benign tissue or to grade treatment response, all calibrated against conventional DWI acquisitions. If a department switches to Zoom-DWI and plugs its measurements into those established thresholds, the systematic bias could push borderline lesions across diagnostic cutoffs and change clinical decisions. The correlation between the two techniques was extremely strong for ADC, with a Pearson correlation coefficient of 0.906, and strong for entropy at 0.776, meaning the two methods track each other well in relative terms. But correlation is not agreement. Bland-Altman analysis, the standard statistical tool for assessing whether two measurement methods can be used interchangeably, revealed consistent systematic ADC deviation between the sequences and wide variation in entropy measurements. The authors’ conclusion is unambiguous: the two sequences cannot share unified quantitative diagnostic cutoffs.

Entropy, interestingly, emerged as the more forgiving metric. Mean entropy values were statistically indistinguishable between the two sequences, at 4.70 for Zoom-DWI and 4.72 for General-DWI, a difference with no statistical significance. This suggests that texture heterogeneity measurements may be more robust to the choice of acquisition technique than ADC values are, an encouraging sign for researchers pursuing radiomics approaches to gastric cancer characterization. However, the wide limits of agreement observed for entropy in the Bland-Altman analysis caution against assuming that individual patient measurements can be swapped freely between sequences, even if group-level means agree.

The study’s practical message for clinics is measured and clear. Zoom-DWI delivers superior visualization of tumor margins, which could improve staging accuracy and help clinicians delineate disease, but it carries higher image noise and produces systematically lower ADC values that are incompatible with existing diagnostic thresholds. The authors therefore recommend that Zoom-DWI serve only as a complementary sequence in gastric cancer MRI, deployed when crisp anatomical delineation is the priority, while conventional DWI remains the backbone for quantitative ADC-based assessments. For patients, the takeaway is that the sharper picture on the radiologist’s screen is real and valuable, but the numbers behind it must be interpreted in the context of the technique that produced them. As targeted diffusion imaging spreads through clinical practice, this study offers a timely reminder that in medical imaging, seeing better and measuring the same are two very different achievements, and that the transition between techniques demands recalibration, not just enthusiasm.

Subject of Research: Comparison of Zoom-DWI and conventional diffusion-weighted MRI for image quality, ADC values, and entropy in gastric cancer imaging

Article Title: Zoom-DWI versus general-DWI in gastric cancer: a comparative evaluation of image quality, apparent diffusion coefficient, and entropy

Article References: Xie, H., Wang, H., Cao, Z., Zhao, M., Chen, D., Guo, Y., Liu, J., Zhang, L., Wang, X., Ran, Q., & Wang, S. (2026). Zoom-DWI versus general-DWI in gastric cancer: a comparative evaluation of image quality, apparent diffusion coefficient, and entropy. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02766-6

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02766-6

Keywords: gastric cancer, diffusion-weighted imaging, Zoom-DWI, apparent diffusion coefficient, MRI, image quality, entropy, echo-planar imaging, radiology, tumor delineation, Bland-Altman analysis, cancer imaging

The article, titled Reading Between the Lines: An Exercise in Clinical Reasoning, appeared online on 13 October 2026 and belongs to a distinctive genre within medical education literature: the blinded case discussion. In this format, one clinician—the discussant—is presented with a real patient’s case piece by piece, without knowing the eventual diagnosis, and is asked to think aloud as the clinical picture unfolds. Everything the discussant says is recorded and analyzed, and the resulting transcript becomes a teaching document in its own right. Fretz served as the blinded discussant for this exercise, while Sanchez analyzed the discussion and Parsons, a recognized scholar of clinical reasoning, reviewed the work and created a figure tracking how diagnostic uncertainty changed over the course of the encounter. The format’s power lies in its honesty: instead of presenting diagnosis as a tidy retrospective narrative, it exposes the false starts, the anchoring on early hypotheses, and the incremental revisions that characterize real clinical thought.

At the heart of the paper is a concept that has been gaining traction in the medical education literature: entropy. Borrowed from information theory, entropy in this context is a measure of uncertainty—of how many plausible explanations remain on the diagnostic table at any given moment. A recent narrative review by Rohlfsen, Shannon, and Parsons in the same journal framed clinical decision-making through the lens of decision theory and entropy, arguing that skilled diagnosticians can be understood as agents who progressively reduce uncertainty by gathering information that discriminates between competing hypotheses. Each test, each examination finding, and each fragment of history either removes entropy, by eliminating possibilities, or fails to do so, leaving the diagnostic space crowded. The new exercise applies this framework to a live case, and Parsons’s figure visualizes that entropy dropping over time as the discussant’s reasoning converges—or fails to converge—on the correct answer.

This quantitative framing matters because it challenges a long-standing assumption in medical training: that expertise is simply a matter of accumulating pattern recognition. Psychologists and education researchers have described how experienced clinicians build rich mental shortcuts, sometimes called illness scripts, that allow instantaneous recognition of familiar diseases. But the references underpinning the new paper point to a more nuanced picture. A study by McQuade and colleagues in the Journal of Hospital Medicine examined the characteristics that differentiate how novices and experts synthesize what researchers call the problem representation—the compressed summary of a patient’s illness that serves as the hinge between raw data and diagnosis. The finding that experts and novices differ not merely in how much they know but in how they structure what they know has profound implications for how medical students should be taught. Related work by Dreicer, Parsons, and colleagues in Perspectives on Medical Education has argued that educators frequently conflate the terms framework and schema, treating them as synonyms when they describe distinct cognitive structures, and that cleaning up this vocabulary is a prerequisite for teaching reasoning well.

While the full text of the exercise is available to journal subscribers, the paper’s reference list reveals the diagnostic territory the case explores, and it is dramatic territory indeed. Among the cited works are reports on pulmonary alveolar proteinosis triggered by immunosuppressive therapy in a transplant recipient, on lymphangitic carcinomatosis in which cancer cells spread through the lymphatic channels of the lung, and on pulmonary tumor embolism, a rare and notoriously elusive condition in which tumor cells obstruct the pulmonary vasculature. These are not the bread-and-butter diagnoses that announce themselves clearly; they are the great mimics of pulmonary medicine, conditions whose presenting complaint—progressive shortness of breath, or dyspnea—can be caused by dozens of more common entities. Springer’s own subject indexing for the article lists dyspnea, diagnostic markers, radiology, and predictive medicine among its related topics, underscoring that the case turns on the interpretation of imaging and the disciplined narrowing of a wide differential diagnosis.

The challenge posed by such cases is precisely why the authors chose the entropy framework. When a patient presents with breathlessness and an abnormal chest image, the initial diagnostic space is enormous: infection, heart failure, clot, malignancy, interstitial lung disease, and many rarer entities all compete. A clinician who orders tests at random will generate data without reducing uncertainty, because much of that data is redundant—what information theorists would call high in entropy overlap. The cited work on entropy removal of medical diagnostics and entropy removal of clinical features, published in Scientific Reports in 2024 and 2025, explores exactly this problem: how to select the sequence of clinical features and tests that maximally shrinks the hypothesis space with each new piece of information. In that light, the expert clinician resembles an efficient search algorithm, and diagnostic excellence becomes a measurable property of the reasoning process rather than an ineffable gift.

The paper also engages with the darker side of diagnostic cognition: the systematic errors to which even experienced physicians succumb. One cited reference, by de Alencar Neto and Santos-Neto in the Journal of General Internal Medicine, examines what the authors call the post hoc pitfall—rethinking how sensitivity and specificity, the statistical properties of diagnostic tests, actually behave in clinical practice when tests are interpreted in sequence rather than in isolation. Another cited work, by Gheihman, Johnson, and Simpkin in Medical Teacher, offers twelve tips for thriving in the face of clinical uncertainty, an acknowledgment that ambiguity is not a failure state to be eliminated but a permanent condition of medical work to be managed. The exercise format makes these hazards visible in real time: readers watch the discussant weigh the possibility of anchoring too early on a plausible but wrong hypothesis, and see how the deliberate articulation of competing explanations serves as a safeguard against premature closure, one of the most frequently documented contributors to diagnostic error.

Why should anyone outside a hospital care about the fine structure of clinical reasoning? The answer lies in the scale of the problem. Diagnostic errors contribute substantially to preventable morbidity and mortality, and most of these errors are cognitive rather than technical—the right test was available, but the wrong conclusion was drawn from it. Teaching programs have traditionally transmitted reasoning skills by osmosis, hoping that trainees would absorb diagnostic acumen from watching skilled mentors. Papers like this one represent a shift toward making the invisible visible: by transcribing and analyzing expert reasoning against a framework of entropy, problem representation, and schema, they give educators concrete material to teach and learners concrete models to imitate. The authors’ explicit division of labor—discussant, analyst, and reasoning scholar—mirrors the collaborative structure that medical education researchers have advocated for turning individual cases into generalizable lessons.

There is also a forward-looking dimension. As artificial intelligence systems move into clinical practice, offering differential diagnoses and flagging imaging findings, a pressing question is whether machines and humans reason in comparable ways. The entropy framework provides a common language: an algorithm that selects the maximally informative next test and a physician who asks the single most discriminating question are, in principle, doing the same thing. Studies of how experts and novices differ in synthesizing problem representations, and of how uncertainty can be quantified and tracked through a diagnostic encounter, will inform how such tools are evaluated and integrated. In the meantime, the humble blinded case discussion remains one of the most powerful instruments available for studying the mind of the diagnostician. Reading between the lines of this exercise, readers find not just the solution to one patient’s puzzle but a demonstration that the process of reaching it can itself be examined, measured, and taught—and that the difference between a good diagnosis and a missed one often lies in the disciplined management of uncertainty, one piece of information at a time.

Subject of Research: Clinical reasoning and diagnostic decision-making using a blinded case exercise and entropy-based analysis of uncertainty

Article Title: Reading Between the Lines: An Exercise in Clinical Reasoning

Article References: Sanchez, A., Parsons, A., & Fretz, A. (2026). Reading Between the Lines: An Exercise in Clinical Reasoning. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10914-5

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10914-5

Keywords: clinical reasoning, diagnosis, diagnostic error, entropy, medical education, dyspnea, pulmonary medicine, problem representation, clinical uncertainty, radiology, Journal of General Internal Medicine, case discussion

Linguistic steganography is the art of hiding secret information inside innocuous-looking text. In a typical covert communication scenario, a sender embeds a hidden message into seemingly normal sentences and transmits the resulting stegotext over a public channel, where the intended recipient recovers the payload. Unlike older modification-based approaches that alter existing documents, generative linguistic steganography embeds information directly during text generation, so that the statistical distribution of the stegotext remains close to that of ordinary language model output. This distribution-preserving property has become a foundational principle of provably secure linguistic steganography, because a message that leaves no statistical fingerprint cannot, in principle, be detected.

The new work targets a specific and increasingly important class of text generators: diffusion language models. Rather than predicting tokens one after another from left to right, as autoregressive models do, diffusion models progressively denoise noisy token states over many inference steps. At each denoising step, the model produces a token-level probability distribution for every position in the sequence. The researchers recognized that this iterative process exposes a whole trajectory of intermediate distributions, each one a potential embedding opportunity with its own entropy characteristics. Previous diffusion-based steganography methods, however, largely ignored this trajectory-level structure.

The team’s motivational study revealed the core problem. Analyzing candidate embedding positions across diffusion trajectories under their default settings, they found that embedding opportunities are highly sparse: the majority of denoising timesteps provide no valid candidates at all, and candidate-rich timesteps occur only occasionally. Existing token-level diffusion steganography methods such as STEAD perform embedding independently within each isolated timestep, which means that when a single step lacks enough compatible candidates, otherwise usable positions are simply wasted. This timestep-local design, the authors argue, is the fundamental bottleneck limiting embedding capacity.

TE-CSA’s answer is an accumulate-then-trigger strategy. Instead of consuming candidates immediately, the framework continuously extracts candidate positions from every denoising timestep and stores them in a global pending pool. When the pool contains at least three compatible positions, the framework constructs an embedding group from the three highest-capacity candidates. Because the same secret bit segment is redundantly embedded across the three positions, the effective capacity of each group is determined by its weakest member, the minimum capacity among the three. This group constraint is central to the paper’s theoretical analysis, in which the authors derive a trajectory-level capacity bound and prove that sorting candidates by capacity and grouping them consecutively attains that bound when the complete retained candidate set is available.

The second key ingredient is entropy-aware filtering. From an information-theoretic standpoint, not all candidate positions are equally valuable. Positions whose token distributions are highly concentrated, meaning the model is already nearly certain which token to place there, carry almost no usable uncertainty and act as bottlenecks: if such a low-capacity position joins an otherwise high-capacity group, it caps the group’s payload. Computing full Shannon entropy for every candidate would be expensive, so the framework uses a min-entropy-inspired approximation based on the maximum token probability, which requires only identifying the most probable token in each local distribution. Candidates whose estimated capacity falls below a threshold are removed before group construction. The strongest filtering configuration, TE-CSA-τ3, retains only positions with capacity of at least 4.

Embedding itself proceeds through distribution-preserving probabilistic sampling. The secret bit sequence is mapped into a continuous value in the interval from 0 to 1 using binary fraction encoding, and the token probability space at each candidate position is partitioned into cumulative intervals. The value falls into one interval, and the corresponding token is sampled under constraints that keep the selection consistent with the original diffusion distribution. Crucially, the authors emphasize that cross-step aggregation and entropy filtering only decide which candidate positions participate in embedding; they do not modify the token probabilities themselves. The security property is therefore inherited from the underlying interval-based embedding mechanism, a point the authors frame as conditional distribution-preserving security rather than unconditional provable security. During extraction, the receiver reconstructs the same candidate sets, filtering decisions, groups, and intervals to recover the message without altering the received text.

The experimental results are striking. Under the default generation setting of length 256, temperature 1.3, and top-p 0.98, the STEAD baseline achieved only 25.30 bits of embedding capacity, while TE-CSA variants reached 469.90 bits without filtering and 470.50 bits with the strongest filtering, an 18.60-fold improvement. The embedding rate rose from 0.09 to 1.84 bits per token, a 20.40-fold gain. Importantly, the improvement does not come from creating new entropy but from converting a far larger fraction of the embedding opportunities already scattered across the diffusion trajectory into effective embedding groups. Runtime overhead remained modest: encoding took 51.53 seconds for TE-CSA-τ3 versus 49.76 seconds for STEAD, because the framework requires no extra denoising iterations or auxiliary neural networks.

Quality and detectability held up as well. Measured by relative perplexity variation, the no-filter variant perturbed generation by 11.20 percent, but entropy-aware filtering cut that figure in half to 5.60 percent while leaving capacity essentially unchanged, validating the claim that low-entropy positions introduce disproportionate distortion for minimal gain. In steganalysis tests against three neural detectors, FCN, R-BiLSTM-C, and BiLSTM-Dense, detection error rates for TE-CSA-τ3 hovered near the 50 percent random-guessing level, at 50.10, 50.06, and 51.38 percent respectively, suggesting the evaluated detectors could barely distinguish stegotext from normal diffusion-generated text. The authors caution that these results characterize practical resistance under the selected detectors, not universal undetectability, and note that LLM-based steganalysis remains a future evaluation frontier.

The study also situates TE-CSA among competing approaches. GTSD, an earlier diffusion-based method, embeds information through prompt mapping and batch selection, with reported rates of roughly 1.56 to 3.94 bits per token depending on configuration, while CascSamp achieves 3.42 bits per token on an autoregressive backbone. TE-CSA reaches a comparable order of magnitude within the more constrained token-level diffusion setting, without requiring pre-shared prompt tables or external mapping resources. The authors acknowledge limitations: the framework depends on trajectories with sufficient candidate diversity, its fixed thresholds may be suboptimal under dynamic generation conditions, and adaptive parameter selection could further improve the capacity-quality-efficiency trade-off. Future directions include extending the approach to other diffusion architectures, longer texts, and multimodal diffusion generation, as well as testing against more advanced steganalyzers.

Subject of Research: High-capacity distribution-preserving linguistic steganography in diffusion language models using trajectory entropy-aware cross-step aggregation

Article Title: TE-CSA: trajectory entropy-aware cross-step aggregation for improving embedding capacity of distribution-preserving linguistic steganography

Article References: Deng, R., Yi, X., Zhang, H., Cao, Y., & Liu, C. (2026). TE-CSA: trajectory entropy-aware cross-step aggregation for improving embedding capacity of distribution-preserving linguistic steganography. Cybersecurity, 9(1), Article 238. https://doi.org/10.1186/s42400-026-00668-2

Image Credits: AI Generated

DOI: 10.1186/s42400-026-00668-2

Keywords: linguistic steganography, diffusion language models, information hiding, entropy, embedding capacity, distribution preservation, cybersecurity, text generation, steganalysis, information theory, covert communication, natural language processing