Our Work

Fix theta between 1/2 and 1. We localize Levinson's method, with Conrey's general operator polynomial Q, to the intervals (T, T+T^theta], for mollifiers of length T^nu with nu less than theta-1/2, by …

Fix theta between 1/2 and 1. We localize Levinson's method, with Conrey's general operator polynomial Q, to the intervals (T, T+T^theta], for mollifiers of length T^nu with nu less than theta-1/2, by running Young's short proof of the mollified second moment with a weight supported on the window. On such windows the method counts distinct sign changes of Hardy's function, with lower density kappa = 1 - log(c)/R, for every Q with Q(x)+Q(1-x) constant. For operator polynomials of degree 201 we certify kappa above 0.7170 nu for nu between 0.0199 and 0.0999. Used as the odd-support input of the Hilbert-parity inequality (Q-S)(N-O) at least 2(N-S)^2, together with Wang's short-interval pair-correlation theorem, this gives a positive proportion of simple critical zeros in (T, T+T^theta] for every fixed theta in [0.534, 1), compared with the previous exponents 0.552, 0.55019 and 0.5458838. At theta = 0.5459 the lower density exceeds 0.017763, about one thousand times the previous bound. Exact rational and ball arithmetic certify all constants, with an independent validated-quadrature replay. The results are asymptotic, use Wang's recent preprint for the pair term, and do not prove the Riemann hypothesis.

Code and publication outputs for Stress testing insurance market stability under climate risk. The model connects hurricane losses, insurance coverage, insurer capital, risk transfer, and institutiona…

Code and publication outputs for Stress testing insurance market stability under climate risk. The model connects hurricane losses, insurance coverage, insurer capital, risk transfer, and institutional financing capacity. Version 1.1.0 includes the finalized financial calculations, reproduction scripts, a synthetic-data demonstration, and checksummed provenance for 106 validated analyses. The full system-level result archive is attached to the matching GitHub release.

This study examines the cognitive processes and decision-making patterns of commercial pilots during high-workload flight scenarios. Using a mixed-methods approach combining eye-tracking, physiologica…

This study examines the cognitive processes and decision-making patterns of commercial pilots during high-workload flight scenarios. Using a mixed-methods approach combining eye-tracking, physiological measurements, and simulator-based performance data, we analyzed decision-making performance in 45 experienced pilots across various challenging conditions. Our findings reveal significant correlations between cognitive workload, attention allocation patterns, and decision quality. Pilots demonstrated adaptive strategies for information prioritization, though experience level showed moderating effects on performance degradation. These results have important implications for pilot training programs, aircraft design, and aviation safety management systems. The study contributes to understanding human factors in aviation safety and provides evidence-based recommendations for enhancing pilot performance under stress.

Se presenta el caso de un hombre de 36 años que consultó por dolor abdominal de 72 h de evolución, iniciado en epigastrio y migrado al hemiabdomen inferior, con náuseas y sensación febril. En la explo…

Se presenta el caso de un hombre de 36 años que consultó por dolor abdominal de 72 h de evolución, iniciado en epigastrio y migrado al hemiabdomen inferior, con náuseas y sensación febril. En la exploración había irritación peritoneal generalizada. Sin radiografía ni tomografía disponibles, se operó con diagnóstico preoperatorio de peritonitis de probable origen apendicular. La laparotomía mostró ~500 mL de líquido purulento y un apéndice retrocecal flemonoso; la discordancia entre el aspecto apendicular y el volumen de contaminación motivó ampliar la exploración, con hallazgo de úlcera gástrica antral perforada (~0,5 cm), reparada con rafia y epiploplastia tipo Graham. La evolución hospitalaria de cuatro días fue favorable. El mensaje educativo es la exploración sistemática ante peritonitis incongruente con el apéndice, sobre todo cuando falta imagen preoperatoria.

Code, synthetic training data and trained models for "Scale-Dependent Limits on Recovering Greenland Ice Loss from Sea Level Fingerprints" (Bao, Coulson, Valencic, Mousavi, Dangendorf, Lloyd and Mitro…

Code, synthetic training data and trained models for "Scale-Dependent Limits on Recovering Greenland Ice Loss from Sea Level Fingerprints" (Bao, Coulson, Valencic, Mousavi, Dangendorf, Lloyd and Mitrovica, submitted to AGU Advances). The direct iterative Fourier neural operator (DIFNO) maps fields between disjoint scattered point sets using direct spectral evaluation with precomputed Vandermonde matrices. A forward DIFNO emulates the elastic sea level equation (Greenland ice-thickness change to sea surface height, SSH, rate), and a physics-informed inverse DIFNO maps SSH back to ice change, constrained by a cycle-consistency loss through the frozen forward model. Files DIFNO-Greenland_code.zip: Jupyter notebooks for training the forward (01_train_forward.ipynb) and inverse (02_train_inverse.ipynb) models, and the model, loss and gradient-surgery source code (src/). Same as the GitHub repository https://github.com/xiyuanbao/DIFNO-Greenland at release time. Requirements are listed in requirements.txt. DIFNO-Greenland_data_*.zip and DIFNO-Greenland_model_*.zip: the data and trained models, split into seven archives of about 350 MB each. Unzip all of them next to the notebooks (unzip '*.zip'); together they form one folder, DIFNO-Greenland_data/, containing- no_smooth/: three synthetic ensembles (single, multi and coastal ice-change families), 10,000 samples each, with ice-thickness change rate at 4,252 Greenland points and the corresponding SSH rate at 4,968 ocean points (m/yr), computed with the sea level equation at spherical harmonic degree 512;- 512full_ref_Gr_case.npz: the reference case, ice-thickness change rate from Simonsen et al. (2021, doi:10.1029/2020GL091216) and its SSH fingerprint on the same points;- 512Gr_oc_coords_full.npz and 512scaled_display_coords_full.pt: point coordinates in longitude and latitude, and in the scaled polar stereographic coordinates used by the models;- _Models/: the four trained models used in the paper (default and coastal forward, default and coastal inverse).README_data.md (in DIFNO-Greenland_data_coastal_coords_ref.zip) describes each file, array shape and unit.

Version 2 Comprehensive Revision: This revised monograph supersedes Version 1. Version 2 substantially expands the interactive Lean 4 formal verification suite (WerracleProof.lean, compiled under Lean…

Version 2 Comprehensive Revision: This revised monograph supersedes Version 1. Version 2 substantially expands the interactive Lean 4 formal verification suite (WerracleProof.lean, compiled under Lean v4.34.1 and Mathlib4 with zero sorry axioms) to include ten machine-verified theorems: (i) explicit algebraic proofs of additive closure, multiplicative ideal absorption, and orthogonal state partitioning over ZMod 9 (ℤ/9ℤ); (ii) 64/128-bit Q16.16 fixed-point arithmetic overflow-safety bounds below 263−1; (iii) formal refutation of constant-output degeneracy via machine-verified escape spectrum and 4-quadrant Observer Horizon sensitivity theorems; and (iv) a parametric Ethereum Virtual Machine (EVM) gas cost model bounding worst-case execution to 22,557 ≤ 24,000 gas. Version 2 also formalizes the two-tier execution architecture linking the 36-iteration CPU edge engine (arXiv:2609.25498, arXiv:2609.30115) with the 12-iteration on-chain EVM reflex kernel, harmonizes all 40-core benchmark tables, reports explicit clean-signal preservation rates (89.60%), and provides a complete open-source replication archive. Contemporary neural inference architectures rely on dense floating-point weight matrices stored in high-bandwidth memory (VRAM), incurring severe memory-wall bottlenecks and preventing native execution inside deterministic, resource-constrained virtual machines such as the Ethereum Virtual Machine (EVM). While formal verification of finite feedforward ReLU networks is well established as NP-complete via SMT and branch-and-bound solvers (e.g., Reluplex, Marabou, α,β-CROWN), verifying termination and arithmetic invariants for unbounded recursive dynamical systems over continuous domains is generally undecidable in the Blum–Shub–Smale model. Here, we present the formal verification and bare-metal empirical validation of WERR (Waves & Errors) and Phase III Orbital Error Dynamics (OED), a non-tensor decision paradigm that procedurally synthesizes non-linear decision boundaries on demand from a 24-byte coordinate seed Θ = (cx, cy, zoom) along the boundary of the Mandelbrot set (∂M). Replication Package Manifest (v2.0): Lean 4 Proof Module: WerracleProof.lean (10 machine-verified theorems compiled under Lean 4 v4.34.1 + Mathlib4, 0 sorry, SHA-256: 35a6a1b75f6056b7a516f849ea2d7113d113569a48a4893474c16bfdba71e862) alongside Verify.lean, lakefile.toml, and lean-toolchain. On-Chain EVM Smart Contracts: WerrMath.sol, Werracle.sol, and WerracleFeeHook.sol. Bare-Metal 40-Core Gauntlet Scripts & Cryptographic Seal: benchmark_oed_40cores.py, ablation_experiment_3arms.py, and OED_40CORE_GAUNTLET_SEAL.json (SHA-256: 94ddaefb17989c9031697779d95f35f7cd5ea3a2bea36ef4c91db1a270061550). TÜRKPATENT Priority Application: TR 2026/016285.

Introdução: A erliquiose monocítica canina é uma enfermidade infecciosa de ampla distribuição causada pela bactéria Ehrlichia canis, cujo diagnó…

Introdução: A erliquiose monocítica canina é uma enfermidade infecciosa de ampla distribuição causada pela bactéria Ehrlichia canis, cujo diagnóstico ainda apresenta limitações relacionadas à sensibilidade e especificidade dos métodos convencionais. A aplicação de ferramentas imunoinformáticas representa uma estratégia promissora para a identificação de biomarcadores com potencial de utilização em testes imunodiagnósticos. Objetivo: Identificar e caracterizar regiões antigênicas das proteínas GP19 e TRP36 de E. canis com potencial aplicação no desenvolvimento de imunodiagnósticos. Metodologia: Foram recuperadas sequências proteicas do banco NCBI e realizadas análises de conservação molecular por alinhamento múltiplo utilizando Clustal Omega. As proteínas foram avaliadas quanto à antigenicidade, acessibilidade de superfície e hidrofilicidade por ferramentas do IEDB, além da caracterização físico-química pelo ProtParam e antigenicidade global pelo VaxiJen. Também foram realizadas predições de ligação ao MHC I e II, modelagem estrutural tridimensional pelo SWISS-MODEL e integração dos resultados para seleção dos epítopos candidatos. Resultados: A GP19 apresentou elevado grau de conservação entre os isolados analisados, enquanto a TRP36 apresentou regiões conservadas intercaladas por segmentos repetitivos variáveis. Ambas as proteínas exibiram regiões com elevado potencial antigênico, hidrofílicas e acessíveis à superfície, compatíveis com epítopos lineares de células B. A integração das análises permitiu identificar peptídeos candidatos com potencial de reconhecimento imunológico e possível aplicação em estratégias diagnósticas. Conclusão: A abordagem imunoinformática permitiu identificar regiões promissoras nas proteínas GP19 e TRP36, fornecendo candidatos para futuras validações experimentais e contribuindo para o desenvolvimento de ferramentas imunodiagnósticas voltadas ao diagnóstico da erliquiose monocítica canina.  

Each of these five categories has many variants. A categorization of the PPP/PSP models together with their aim characteristics is shown in table below. While the spectrum of models shown in the table…

Each of these five categories has many variants. A categorization of the PPP/PSP models together with their aim characteristics is shown in table below. While the spectrum of models shown in the table are possible as individual options, combinations are also possible such as, a lease or (partial) privatization contract for existing facilities which incorporates provisions for expansion through Build-OperateTransfer. In fact, many PPP projects of recent times are of combination type.