Papers

The expansion of global electricity distribution systems necessitates the deployment of massive infrastructure. Assessing its implications from a spatial and material perspective requires an understanding of the core drivers of a distribution grid configuration. Our model samples substation locations using a non-linear relationship with population density and constructs a proxy network applying the Kruskal algorithm. This streamlined approach generates a proxy grid layout at the local scale and provides reasonable aggregate estimates of the total network length at the national scale. Using highly granular population data, this local model reveals a connection between population spread and distribution grid, which appears to persist at larger scales. Potentially driven by the emergent properties of population scaling laws, the aggregated network characteristics appear to be well described by multivariate power laws on aggregated population and area. Benchmarked against reported aggregate data for 35 countries, these results provide new multi-scale tools for characterizing electrical infrastructure and reveal key determinants of distribution grid extent. Combining network length estimates with material intensity data, we derive country-scale inventories of copper invested in medium-voltage lines, bounded by aerial and underground cabling assumptions. The same scaling law directly yields a macro-level proxy for copper mass as a function of population and area, extending the framework toward prospective assessments of material demand in electricity grid expansion.

In order to achieve its ambitious luminosity target, the SuperKEKB collider must achieve and sustain high beam currents on the order of Ampere in its storage rings. This requires continuous top-up injection and operation with a two-bunch injection scheme, injecting two 96 ns spaced bunches in a single injection cycle. An important input for tuning the injection beam is the position reading from a dedicated beam position monitor (BPM), located after the septum magnets, slightly upstream of where the injected and stored beams converge. Previously, the readout electronics used for these special BPMs were not capable of independent measurement of both bunches in the two-bunch injection mode and modification of the concerned devices and their firmware was not feasible. The opportunity was taken to develop a new readout device based on the AMD/Xilinx RF System on a Chip (RFSoC) platform with the goal of not only providing a sufficiently flexible and performant readout solution for the concerned BPMs, but also to evaluate and gain experience with the platform for beam monitor electronics applications. This paper is concerned with the details of this development as well as evaluation and operation of the developed device.

Fluidic Shaping is a novel method for fabrication of optical components based on the equilibrium state of liquid volumes in neutral buoyancy, subjected to geometrical constraints. The underlying physics of this method is described by a highly nonlinear partial differential equation with Dirichlet boundary conditions and an integral constraint. To date, useful solutions for such optical liquid surfaces could be obtained analytically only for the linearized equations and only on circular or elliptical domains. A numerical solution for the non-linear equation was suggested, but only for the axi-symmetric case. Such solutions are, however, insufficient as they do not capture the full range of optical surfaces. Arbitrary domains offer an important degree of freedom for creating complex optical surfaces, and the nonlinear terms are essential for high quality solutions. Moreover, in the context of optics, it is not sufficient to resolve the shape of the surface, and it is essential to obtain accurate solutions for its curvature, which governs its optical properties. We here present the theoretical foundation for the Fluidic Shaping method over arbitrary domains, and the development of a high order (quintic) finite element numerical solver, capable of accurately resolving the topography and curvature of liquid interfaces on arbitrary domains. The code is based on reduced quintic finite elements, which we have modified to capture curved boundaries. We compare the results against low order finite elements and non-deformed high order elements, demonstrating the importance of high order approximations of both the solution and the domain. We also show the usability of the code for the prediction of optical surfaces derived from complex boundary conditions.

In this paper, we propose a domain decomposition dynamical low-rank method to solve high-dimensional radiative transfer problems and similar kinetic equations. The algorithm uses a separate low-rank approximation on each spatial subdomain, which means that, for a given accuracy, we can often use a smaller overall rank compared to classic dynamical low-rank methods. In particular, we can solve problems with point sources efficiently, that for classic algorithms require almost full rank. Our algorithm only transfers boundary data between subdomains and is thus very attractive for distributed memory parallelization, where classic dynamical low-rank algorithms suffer from global data dependency. We demonstrate the efficiency of our algorithm by a number of challenging test examples that have both very optical thin and thick regions.

The traditional limitations of neural networks in reliably generalizing beyond the convex hulls of their training data present a significant problem for computational physics, in which one often wishes to solve PDEs in regimes far beyond anything which can be experimentally or analytically validated. In this paper, we show how it is possible to circumvent these limitations by constructing formally-verified neural network solvers for PDEs, with rigorous convergence, stability, and conservation properties, whose correctness can therefore be guaranteed even in extrapolatory regimes. By using the method of characteristics to predict the analytical properties of PDE solutions a priori (even in regions arbitrarily far from the training domain), we show how it is possible to construct rigorous extrapolatory bounds on the worst-case L^inf errors of shallow neural network approximations. Then, by decomposing PDE solutions into compositions of simpler functions, we show how it is possible to compose these shallow neural networks together to form deep architectures, based on ideas from compositional deep learning, in which the large L^inf errors in the approximations have been suppressed. The resulting framework, called BEACONS (Bounded-Error, Algebraically-COmposable Neural Solvers), comprises both an automatic code-generator for the neural solvers themselves, as well as a bespoke automated theorem-proving system for producing machine-checkable certificates of correctness. We apply the framework to a variety of linear and non-linear PDEs, including the linear advection and inviscid Burgers' equations, as well as the full compressible Euler equations, in both 1D and 2D, and illustrate how BEACONS architectures are able to extrapolate solutions far beyond the training data in a reliable and bounded way. Various advantages of the approach over the classical PINN approach are discussed.

Laser wakefield theories rely on the laser envelope function, which is radially symmetric, and predict zero transverse momentum for the electrons along the propagation axis. Exact description of laser wakefields, beyond the envelope approximation, requires a more general formula for the Lorentz force acting on the electrons. Here we present a fundamental approach to express the transverse momentum of an electron crossing the laser pulse, and we show that an exact analytical formula can be derived for the non-zero transverse momentum of electrons initially lying along the axis of symmetry. The results outlined here shed light on the details of the electron motion inside an intense laser pulse and explain the strong wakefield asymmetry observed in simulations.

This paper presents experimental validation of the homogenized harmonic balance finite element method (HomHBFEM), which we have developed as a dedicated simulation technique for magnets with fast excitation cycles, in particular the fast corrector (FC) magnets for PETRA IV at DESY. The HomHBFEM allows efficient three-dimensional nonlinear eddy-current simulations of laminated magnets at elevated frequencies with a relatively coarse finite element (FE) mesh and without computationally expensive time-stepping. This is achieved by combining a frequency-domain-based homogenization technique with the harmonic balance FE method. The simulation results for the magnetic flux density along the axis of the FC magnets as a function of frequency and the resulting integrated transfer function (ITF) are compared to Hall probe and search coil measurements of the first prototype FC magnet for PETRA IV. A good agreement between simulated and measured ITFs is achieved for excitation frequencies from 10 Hz to 10 kHz.

The excavation of the Hyper-Kamiokande cavern, 600 m underground, is complete. Measuring 69 m in diameter and 94 m in height, it is among the world's largest rock caverns. A vertically oriented, dome-capped cylindrical design was chosen to optimize cost and performance. The geometry, combined with substantial overburden, posed major engineering challenges. This paper outlines the underground works, main cavern design, excavation plan, and the evolution of support design and construction methods during excavation, namely the information-based (observational) design and construction approach.

Optoacoustic imaging technologies require fast and accurate signal pre-processing algorithms to enable widespread deployment in clinical and home-care settings. However, they still rely on the Discrete Fourier Transform (DFT) as the default tool for essential signal-conditioning operations, which imposes hard limits on both execution speed and signal-retrieval accuracy. Here, we present a new transform whose building blocks are directly inspired by the physics of optoacoustic signal generation. We compared its performance with the DFT and other classical transforms on common signal-processing tasks using both simulations and experimental datasets. Our results indicate that the proposed transform not only sets a new lower bound on computational complexity relative to the DFT, but also substantially outperforms classical transforms on basic signal-processing operations in terms of accuracy. We expect this transform to catalyze broader adoption of optoacoustic methods.

Valley splitting in strained Si/SiGe quantum wells is a central parameter for silicon spin qubits and is commonly described with envelope-function and effective-mass theories. These models provide a computationally efficient continuum description and have been shown to agree well with atomistic approaches when the confinement potential is slowly varying on the lattice scale. In modern Si/SiGe heterostructures with atomically sharp interfaces and engineered Ge concentration profiles, however, the slowly varying potential approximation underlying conventional (local) envelope-function theory is challenged. We formulate an exact multi-valley envelope-function model by combining Burt-Foreman-type envelope-function theory, which does not rely on the assumption of a slowly varying potential, with a valley-sector decomposition of the Brillouin zone. This construction enforces band-limited envelopes, which satisfy a set of coupled integro-differential equations with a non-local potential energy operator. Using degenerate perturbation theory, we derive the intervalley coupling matrix element within this non-local model and prove that it is strictly invariant under global shifts of the confinement potential (choice of reference energy). We then show that the conventional local envelope model generically violates this invariance due to spectral leakage between valley sectors, leading to an unphysical energy-reference dependence of the intervalley coupling. The resulting ambiguity is quantified by numerical simulations of various engineered Si/SiGe heterostructures. Finally, we propose a simple spectrally filtered local approximation that restores the energy-reference invariance exactly and provides a good approximation to the exact non-local theory.

The El Niño Southern Oscillation (ENSO) is the dominant driver of interannual global climate variability and can lead to extreme weather events such as droughts or flooding. Recently, we have developed several statistical approaches for early ENSO forecasting, in particular, its El Niño phase. The climate network-based approach allows forecasting the onset of an El Niño event or its absence about 1 year ahead [1]. The complexity-based approach allows additionally to forecast the magnitude of an upcoming El Niño event in the calendar year before the onset [2]. Additionally, we have developed methods for forecasting the type (Eastern Pacific or Central Pacific) of an El Niño [3] and for probabilistic forecasting of La Niña and neutral events [4], also by the end of the calendar year before the event. Here we present the forecasts of these methods for 2026. The climate network and the complexity-based approach do not provide concurring signals for this year. The combined forecast indicates that a neutral event is more likely than an El Niño. If an El Niño develops in 2026, the complexity-based approach predicts a weaker event with a magnitude of $0.84\pm0.36${\deg}C.

Lensless digital holographic microscopy (LDHM) relies on interference between an unscattered reference wave and a weakly scattered object wave - an assumption that rapidly fails in turbid samples under multiple scattering. To overcome this limitation, we present near-infrared LDHM (NIR-LDHM), a in-line holographic platform that operates up to the silicon cutoff (~1100 nm) using a conventional board-level CMOS sensor designed for visible (VIS) imaging. Using tissue-mimicking milk scattering layers and calibrated resolution targets, we quantify reconstruction performance versus wavelength, scattering strength, and sample-sensor distance. NIR-LDHM maintains resolvable features through scattering layers up to ~1.4 mm, whereas visible regime fails to resolve features below ~350 um. Importantly, despite a detector quantum efficiency of only 0.19% at 1100 nm, robust reconstructions are obtained under low-photon-budget conditions. We further identify advantageous mechanism by which increasing the sample-sensor distance from ~3 to 12 mm improves lateral resolution by twofold under strong scattering. Finally, we demonstrate wide-field, label-free amplitude-phase imaging of uncleared mouse tissues, resolving internal structure in brain and liver slices up to ~250 um and ~60 um, respectively. By extending lensless complex-field microscopy into strongly scattering regimes with minimal hardware changes, this work has relevance to computational imaging through complex media and biophotonics.

Regenerative amplifier x-ray free-electron lasers constitute alternative configurations for fourth generation light sources which are typically operated in self-amplified spontaneous emission (SASE) mode. However, SASE exhibits relatively large fluctuations in the power and spectral properties from shot to shot, whereas XRAFELs provide a much more stable source of hard x-ray photons. Here, we study the performance of an x-ray cavity under consideration at the Stanford Linear Accelerator Facility using both uniform and step-tapered undulators with either transmissive or hole out-coupling.

We report on a novel dual-structure scaffold fabricated using two-photon polymerisation (2PP), integrating electrically conductive and non-conductive regions within a single architecture, targeting neural tissue regeneration. Poly(ethylene glycol) diacrylate (PEGDA) was combined with 20 nm gold nanoparticles to create a conductive microstructure, encapsulated within a biocompatible PEGDA lattice designed to support neuronal growth possibility after neural injuries. Electrical resistance measurements confirmed the feasibility of integrating conductive pathways into precision-fabricated microarchitectures.

We present and test a recursive regularised lattice Boltzmann method for incompressible magnetohydrodynamic (MHD) flows. The approach is based on a double-distribution formulation, in which the magnetic field is evolved using a standard BGK lattice Boltzmann scheme, while the fluid solver is enhanced through a Hermite-based recursive regularisation of the non-equilibrium moments. The method exploits a fourth-order Hermite expansion of the equilibrium distribution on the D2Q9 lattice, allowing higher-order isotropy to be retained while selectively filtering spurious non-hydrodynamic contributions. The regularisation procedure reconstructs the non-equilibrium distribution from physically consistent Hermite coefficients, avoiding explicit evaluation of velocity gradients. The resulting scheme preserves the correct incompressible MHD limit, improves numerical stability at low viscosities, and reduces lattice-dependent artefacts. The proposed formulation provides a robust and versatile framework for MHD simulations and offers a systematic route for extending regularised lattice Boltzmann methods to coupled multiphysics systems.

We study the problem of the boundary conditions in the numerical simulation of closed and open quantum systems, described by a Schrödinger equation. On one hand, we show that a closed quantum system is defined by local boundary conditions. On the other hand, we argue that, because of the uncertainty principle, no local boundary condition can be defined for open quantum systems. For this reason plane waves or wave packet trains cannot be simulated on a finite numerical lattice with the usual procedures. We suggest a method that avoids these difficulties by using only a small numerical lattice and maintains the correspondence with the physical picture, in which the incident and scattered waves may be infinitely extended.

Machine learning models are increasingly applied in materials science, yet their predictive power is often constrained by data scarcity. Here, we show that accurate predictions can be achieved, even with a limited number of training examples, provided the dataset is compact and and grounded in physically relevant quantities. By combining density functional theory calculations with a machine-learning framework, we construct accurate descriptor-based models to predict the formation energies of doped lepidocrocite TiO$_2$ monolayers. The predictive accuracy of machine-learning models was first evaluated for single-dopant Pt configurations, demonstrating that the selected structural and chemical descriptors reliably capture the key factors governing dopant stability. Chemical transferability is then examined by extending the dataset to include Ag-doped configurations. Predictive accuracy improved systematically as additional Ag-doped data points were included in the training, while the performance of Pt remains robust. These results highlight the potential of small and well-curated datasets combined with physically informed descriptors to enable not only accurate but also chemically transferable machine-learning-driven screening in doped TiO$_2$ monolayer.

A major challenge in open quantum dynamics is the inclusion of Matsubara-decay terms in the memory kernel, which arise from the quantum-Boltzmann delocalisation of the bath modes. This delocalisation can be quantified by the radius of gyration squared ${\mathcal R}^2(\omega)$ of the imaginary-time Feynman paths of the bath modes as a function of the frequency $\omega$. In a Hierarchical Equations of Motion (HEOM) calculation with a Debye–Drude spectral density, ${\mathcal R}^2(\omega)$ is the only quantity that is treated approximately (assuming convergence with respect to hierarchy depth). Here, we show that the well-known Ishizaki–Tanimura correction is equivalent to separating smooth from `Brownian' contributions to ${\mathcal R}^2(\omega)$, and that modifying the correction leads to a more efficient HEOM in the case of fast baths. We also develop a simple `A4' adaptation of the `AAA' (Adaptive Antoulas–Anderson) algorithm in order to fit ${\mathcal R}^2(\omega)$ to a sum over poles, which results in an extremely efficient implementation of the standard HEOM method at low temperatures.

Purpose: This study aims to investigate the use of dynamic MRI to assess abdominal wall biomechanics before and after hernia surgery, considering that such evaluations can enhance our understanding of physiopathology and contribute to reducing recurrence rates. Methods: Patients were assessed using dynamic MRI in axial and sagittal planes while performing exercises (breathing, coughing, Valsalva) before and after their abdominal hernia surgery with mesh placement. Rectus and lateral muscles, linea alba, viscera area, defect dimensions and hernia sac were contoured with semi-automatic process to quantify the abdominal wall biomechanical temporal modifications. Results: This study enrolled 11 patients. During coughing, the axial area of the hernia sac increased by 128.4 $\pm$ 199.2%. The sac increased similarly in axial and sagittal planes during Valsalva. Post-surgical evaluations showed a 26% reduction in inter-recti distance and a lengthening of all muscles (p $\le$ 0.05). The post-operative rectus abdominis thickness change was negatively correlated with defect width during breathing (p $\le$ 0.05). The largest change in linea alba displacement was observed in the surgical site (p = 0.07). Post-operatively, lateral muscles had a larger inward displacement during Valsalva (p $\le$ 0.05). Rectus abdominis had a larger outward displacement during breathing (p = 0.09), reduced with the mesh size (p $\le$ 0.05). A large inter-individual variability was observed. Conclusion: Using a semi-automatic methodology, an in-depth analysis of the biomechanics of the abdominal wall was conducted, highlighting the importance of a patient-specific assessment. A broader study and consideration of recurrence would subsequently complete this methodological work.

Highly stretched capillary jets produced by gravity are central to drop generation, micro-thread formation, and extensional-rheometry concepts. For Newtonian fluids, the transition from steady jetting to self-excited oscillations in a gravitationally stretched jet is predicted accurately by one-dimensional slender-jet equations that retain the exact interfacial curvature and admit a global eigenvalue analysis Rubio-Rubio et al. 2013. Separately, weakly viscoelastic jets governed by Oldroyd–B/Giesekus constitutive laws exhibit elastocapillary regimes and beads-on-a-string dynamics that are well captured by one-dimensional free-surface models Ardekani et al. 2010. Here we formulate a unified one-dimensional model for gravitationally stretched viscoelastic jets, combining full-curvature capillarity with a Giesekus stress closure, and we analyse its global linear stability on spatially developing base states. We first benchmark the Newtonian limit, reproducing marginal spectra and base-flow profiles, and then quantify how elasticity shifts the critical jetting–dripping boundary by tracking the leading global Hopf eigenpair across the rheological parametric space. For experimentally relevant moderate elasticity, characterised by order-unity Deborah numbers, polymeric tension modifies both the critical Weber number and the selected oscillation frequency, and endogeneity decompositions reveal that marginality results from a balance between capillary/kinematic contributions and an additional elastic-stress feedback pathway. To interpret and predict the onset mechanism, we compute wavemakers and receptivity/structural-sensitivity fields from direct–adjoint eigenfunctions, showing that viscoelasticity broadens the sensitivity region downstream while the adjoint remains strongly localized near the inlet, thereby identifying the near-nozzle region as the dominant receptive location.

We explain the achievements that were awarded 2022 Nobel Prize in Physics, as well as the preceding and the later developments. The main notions and historic cornerstones of Bell inequalities, the related researches on quantum entanglement are reviewed, and the key physical ideas are emphasized. Among the early work, C. S. Wu's contributions using polarization-entangled photons from electron-positron annihilation are introduced.

Zernike phase-contrast imaging is unique among imaging techniques in that it enables the upper limit of Fisher information allowed by quantum mechanics. Here we show that, in a departure from an ideal setting, using an incident beam of finite width, and a $\pi/2$ phase plate having a finite cutoff, the technique can still deliver $>95\%$ of the quantum limit. We point out that the Zernike method is, in principle, applicable to any incident beam. As an example, we sketch an approximate implementation of the method for an incident speckle beam, and show that it too can deliver $>95\%$ of the quantum limit.

We present a two-well terahertz (THz) quantum cascade laser designed for high temperature operation based on the InGaAs/AlInAs material system. The lighter effective mass and higher energy barriers increase the gain at high temperatures (T > 150K). When processed in copper-based double metal waveguides the devices show laser action up to a maximum operating temperature of 188K with a maximum current density of 1.4kA/cm$^2$. The low Joule heating due to reduced active region thickness and low electrical bias allows operation at 10% duty cycle up to a temperature of 170K.

In a recent arXiv posting [1] I reported the result of an experiment: asking Perplexity.ai to compare three items concerning (ordinary) Gamma Ray Burts (GRBs): the data, the standard paradigm(s) and the "Cannonball" (CB) model. Here I ask the same URL to extend this comparison to long–lasting GRBs, binary Neutron-Star mergers and their associated short–hard GRBs, low–luminosity GRBs, X–ray flashes, X–ray transients, and non–solar cosmic rays. The results of this experiment are enlightening but worrisome. Except for this abstract, two footnotes and two other references to standard [2] and CB-model [3] articles and talks, all of what follows is, verbatim, what the cited AI "opines".

Laser-assisted Atom Probe Tomography (LA-APT) has demonstrated a unique potential for the study of the 3D distribution of atomic species in semiconductor materials and devices, and in a growing list of inorganic non-metallic solids. A crucial and often underestimated issue with APT is its accuracy in compositional measurements of non-metallic systems. This work introduces the principles of APT as an experimental technique, recalling the aspects potentially leading to compositional biases and underlining in particular the role of the surface electric field in governing the different physical-chemical phenomena that enable the measurement. It reviews the possible mechanisms of specific losses, as well as the methods for assessing a compositional bias and proposing possible correction methods. Finally, it establishes a state of the art on compositional biases in APT of non-metallic materials, on the basis of which it will be possible to conclude on specific recommendations for best practices, and the perspective of application of APT to new materials.

Sonification – the mapping of data to non-speech audio – offers an underexplored channel for representing complex dynamical systems. We treat El Niño-Southern Oscillation (ENSO), a canonical example of low-dimensional climate chaos, as a test case for culturally-situated sonification evaluated through complex systems diagnostics. Using parameter-mapping sonification of the Niño 3.4 sea surface temperature anomaly index (1870–2024), we encode ENSO variability into two traditional Javanese gamelan pentatonic systems (pelog and slendro) across four composition strategies, then analyze the resulting audio as trajectories in a two-dimensional acoustic phase space. Recurrence-based diagnostics, convex hull geometry, and coupling analysis reveal that the sonification pipeline preserves key dynamical signatures: alternating modes produce the highest trajectory recurrence rates, echoing ENSO's quasi-periodicity; layered polyphonic modes explore the broadest phase space regions; and the two scale families induce qualitatively distinct coupling regimes between spectral brightness and energy – predominantly anti-phase in pelog but near-independent in slendro. Phase space trajectory analysis provides a rigorous geometric framework for comparing sonification designs within a complex systems context. Perceptual validation remains necessary; we contribute the dynamical systems methodology for evaluating such mappings.

We study relativistic twisted (orbital-angular-momentum) states of a massive charged particle propagating through an axially symmetric, longitudinally inhomogeneous solenoid field and a co-directed accelerating or decelerating electric field. Starting from the Dirac equation and using controlled spinless and paraxial approximations, we show that the transverse envelope obeys an effective nonstationary Schrödinger equation governed by a Caldirola–Kanai Hamiltonian. The longitudinal energy gain or loss encoded in $f(z)=[E_0-V(z)]^2-m^2$ generates an effective gain or damping rate $\widetilde{\gamma}(z)=\partial_z f(z)/[2f(z)]$ and a $z$-dependent oscillator frequency $\widetilde{\omega}(z)=p_0\Omega(z)/\sqrt{f(z)}$. Exploiting the Ermakov mapping (unitary equivalence of Caldirola–Kanai systems), we obtain a closed-form propagated twisted wave function by transforming the stationary Landau basis. The transverse evolution is controlled by a single scaling function $b(z)$ that satisfies a generalized Ermakov–Pinney equation with coefficients determined by $E_z(z)$ and $B_z(z)$. In the limiting cases of uniform acceleration with $B_z=0$ and of solenoid focusing with negligible acceleration, our solution reduces to previously known analytic results, providing a direct bridge to established models.

A comparison of the ab initio calculations using the i-DMFT Method by Di Liu et al. (2025) with benchmark potential curves for three HX(F,Cl,Br) halides shows their inaccuracy in the domain around equilibrium - they do not reproduce quantitatively the results of the Born-Oppenheimer approximation - and also they predict a qualitatively wrong behavior in the Van der Waals region of large distances, thus, contradict the multipole expansion.

Coarse-grained molecular dynamics often sacrifices accuracy and transferability for computational efficiency, but the use of machine learned potentials is helping coarse-grained models attain performance on par with atomistic molecular dynamics. Nevertheless, developing representations of the coarse-grained potential energy surface faces severe scaling challenges due to the extreme data demands of widely used "bottom-up" coarse-graining objectives. In this work, we show that mean force matching, a strategy for training thermodynamically consistent coarse-grained models, requires 50x fewer training samples and 87% less total atomistic simulation time, while obtaining better accuracy on the potential of mean force for unseen proteins compared to other commonly used objectives. By systematically removing noise from the objective function, we demonstrate that it is possible to scale machine learning architectures for coarse-graining, enabling highly accurate and transferable models. We show the advantages of mean force matching both theoretically and through exhaustive benchmarking using thermodynamic consistency as the primary metric of accuracy.

The performance of high-precision cold-atom interferometers is often limited by imperfections in the Raman laser fields. We present a reproducible framework for robust Raman mirror-pulse design and compare Krotov, GRAPE, and CRAB under a common normalized peak-amplitude limit of 3.0. The effective two-level model uses a 25-member detuning–amplitude ensemble and a dense out-of-sample grid. In the fixed-budget study, GRAPE attained a terminal ensemble error of $1.224\times10^{-2}$ and projected Krotov attained $1.243\times10^{-2}$; Krotov occupied a slightly larger $P_e\ge0.9$ grid fraction (0.177 versus 0.170). The best of five CRAB seeds gave $3.081\times10^{-2}$. In the interferometer calculation, GRAPE produced the largest contrast, $0.582\pm0.019$, while Krotov and the selected CRAB pulse gave $0.454\pm0.019$ and $0.439\pm0.023$, respectively. These results establish a reproducible comparison of trade-offs, rather than universal superiority of any one optimizer.