Our Work

In 2020, the ICSP — representing microbiologists worldwide — rejected a SeqCode-like proposal by a clear majority. The ICSP specifically rejected the idea of allowing DNA sequences to be u…

In 2020, the ICSP — representing microbiologists worldwide — rejected a SeqCode-like proposal by a clear majority. The ICSP specifically rejected the idea of allowing DNA sequences to be used as nomenclatural types of taxa with a validly published name. It did so for a number of reasons. This rejection applied not only to gene sequences, but also to genome sequences as nomenclatural types of taxa with a validly published name. The SeqCode contravenes the ICNP by design because it implements the proposal that was rejected by the ICSP. The SeqCode deems names to be 'validly published' even if they are based on a sequence as a nomenclatural type. This means that every name 'validly published' under the SeqCode creates an additional logical contradiction with the ICNP. Furthermore, the SeqCode was derived from the ICNP without the ICSP's approval. This document explores the extent to which the SeqCode literally copies the ICNP, raising additional ethical questions.

Coarse-grained snapshots of stochastically forced two-dimensional incompressible turbulence on the periodic domain (0, 2π)², used as the unresolved case (E2) in "Only Linear Constraints Surviv…

Coarse-grained snapshots of stochastically forced two-dimensional incompressible turbulence on the periodic domain (0, 2π)², used as the unresolved case (E2) in "Only Linear Constraints Survive Coarse-Graining: Evaluating Physics-Constrained Neural Operators on Stochastically Forced Turbulence" (NeurIPS 2026 Workshop on AI for Stochastic Dynamics). Five realisations were integrated with Dedalus 3.0.4 at 1024² to t = 2000, with viscosity ν = 9.3×10⁻⁶, linear drag α = 3×10⁻³, and an Ornstein–Uhlenbeck forcing confined to |k| ∈ [56, 72] with injection rate ε = 1.71×10⁻⁵ and correlation time τ = 0.3, at a solver step of 8×10⁻³. Data before t = 924.8 (spin-up) are discarded. Fields are written at 64² every 0.08 time units by a sharp spectral cutoff at |kₓ|, |k_y| ≤ 31, placed below the forcing band, so that the forcing is entirely subgrid on the coarse field; every fifth snapshot is retained, giving a model step of Δt = 0.4 and 2,688 consecutive snapshot pairs per realisation. Channels are the two velocity components and pressure. Simulation seeds 1234–1236 are the training realisations, 1237 validation and 1238 test. The record contains the coarse arrays for the five realisations, plus the normalisation statistics computed from the training inputs. The solver configuration is at https://github.com/m-groom/NS2D and the preparation script (data_utils/prepare_ns2d_stochastic.py), training and evaluation code at https://github.com/m-groom/physics-constrained-neural-operators.

Dependencies AMReX: 26.10 pybind11: 3.0.4 Python: 3.11 - 3.14 What's Changed Fix Sphinx Stubs & Autogen by @ax3l in https://github.com/AMReX-Codes/pyamrex/pull/618 PODVector from_array by @ax3l…

Dependencies AMReX: 26.10 pybind11: 3.0.4 Python: 3.11 - 3.14 What's Changed Fix Sphinx Stubs & Autogen by @ax3l in https://github.com/AMReX-Codes/pyamrex/pull/618 PODVector from_array by @ax3l in https://github.com/AMReX-Codes/pyamrex/pull/580 Release: pyAMReX 26.10 by @EZoni in https://github.com/AMReX-Codes/pyamrex/pull/620 Full Changelog: https://github.com/AMReX-Codes/pyamrex/compare/26.09...26.10

For k > 0 let S(k) = Σ_p 1/(p (log p)^k). We prove that S is real-analytic and strictly convex on (0, ∞) and has a unique minimizer k_ℙ, with three equivalent characterizations: variati…

For k > 0 let S(k) = Σ_p 1/(p (log p)^k). We prove that S is real-analytic and strictly convex on (0, ∞) and has a unique minimizer k_ℙ, with three equivalent characterizations: variational, as an exact balance between the negative atom at p = 2 and the primes p ≥ 3, and as an exponential-tilt centering of log log p. A prime-zeta Mellin representation gives k_ℙ = 1.60499334177498705010… to 109 significant digits, verified by three independent high-precision runs. We prove monotone convergence of truncated minimizers, embed k_ℙ in a strictly decreasing critical curve k_ℙ(s) with an explicit two-prime asymptotic, give a criterion for filtered prime families, and compare with the all-integers minimizer k_ℕ = 2.48810101048… The constants k_ℙ and k_ℕ were denoted k₀ and k₀^E in earlier work of the author. Files: the paper (PDF and DOCX); kP_high_precision.py, the Python/mpmath script that computes k_ℙ and k_ℕ and performs the Diophantine tests; and kP_log_110.json, the log with the 109 verified digits and all run parameters.

This poster presents a multiannual assessment of frontal retreat of ocean-contact glacier fronts on King George Island, Antarctic Peninsula, between 2016 and 2023. From 73 glacier fronts identified on…

This poster presents a multiannual assessment of frontal retreat of ocean-contact glacier fronts on King George Island, Antarctic Peninsula, between 2016 and 2023. From 73 glacier fronts identified on the island, 42 ocean-contact glacier fronts were analyzed, including 15 marine-terminating and 27 mixed marine–land fronts. Austral summer optical satellite imagery from Landsat-8, Sentinel-2 and PlanetScope was used to delineate glacier-front positions, complemented by the GLIMS glacier inventory for front correction and classification. Consecutive-year comparisons were performed to quantify annual and cumulative frontal retreat. The analyzed glacier sample declined from approximately 914.25 km² to 873.73 km² between 2016 and 2023, with a cumulative frontal retreat of approximately 40.52 km² (4.43%). The largest interannual retreat occurred during 2020–2021, reaching approximately 15.93 km². The results demonstrate the usefulness of multi-source optical satellite imagery for monitoring glacier-front evolution in remote Antarctic environments.Presented at the 12th SCAR Open Science Conference (SCAR OSC 2026), Oslo, Norway, 10–14 August 2026. Poster ID: 140.

    Traditionally, Maxwell-Boltzmann entropy is S = -k ln(number of microstates). The microstates which represent Etotal distributed over N particles all have the same weight.&nbsp…

    Traditionally, Maxwell-Boltzmann entropy is S = -k ln(number of microstates). The microstates which represent Etotal distributed over N particles all have the same weight.  -kln(number of microstates), which is called entropy, then seems to be a main conceptual notion of equilibrium. In the case of the Tsallis distribution, however, one does not have equal weight microstates and so it is not clear what function one should maximize a priori.    In (1), we introduced a stability method which yields both the Maxwell-Boltzmann and Tsallis distributions. This approach is based on a function Sum  over i p(ei) F(p(ei)) with p(ei) unnormalized and F(p(ei))=-ei/T such that p(ei) -> p(ei)+dp(ei) induces the same changes in the function as it does in the constraints Sum over i p(ei)=1 and Sum over i ei p(ei)= Eave. There is no notion of maximiing the ln of the number of microstates or any sense of minimal information, at least explicitly.     In other preprints (2), we argued that one may obtain the Maxwell-Boltzmann and Tsallis distributions by considering the simplest conditions in e for p(ei)p(ej)=p(ek)p(el), namely ei+ej = ek+el and ei*ej = ek*el. Here we argue that these simple conditions represent the minimal information in ei associated with product probabilities. We suggest that one may think only in terms of minimal e information for p(ei)p(ej) and not worry about microstates at all. We try to show why this approach is equivalent to the stability one.  

A one-dimensional model of a LiFePO4 half cell (lithium foil, separator, porous cathode) with dilute-solution electrolyte transport, electronic conduction and Butler-Volmer insertion kinetics, solved …

A one-dimensional model of a LiFePO4 half cell (lithium foil, separator, porous cathode) with dilute-solution electrolyte transport, electronic conduction and Butler-Volmer insertion kinetics, solved by finite volumes, backward Euler and Newton's method with the BAND block-tridiagonal solver. It runs constant-current, constant-voltage and rest protocols. Fortran, C++ and Python implementations read the same input file.

Perfect exclusion of quantum states can emerge under collective measurements even when it is impossible on a single copy. We establish the corresponding finite-copy activation law for quantum channels…

Perfect exclusion of quantum states can emerge under collective measurements even when it is impossible on a single copy. We establish the corresponding finite-copy activation law for quantum channels using the Doeblin coefficient of \(n\) channel uses. For every qubit channel, the coefficient vanishes after exactly \(\lceil \ln 2/\ln(1+r^\star)\rceil\) uses, where \(r^\star\) is the largest ratio of Schmidt coefficients among vectors in the kernel of the channel's Choi operator; it remains nonzero for all \(n\) when that kernel contains no entangled vector. The two directions admit closed-form certificates: above the threshold, a loop intervention consisting of one diagonal phase gate and local unitaries; below it, a diagonal dual witness. Classical--quantum channels recover the Pusey--Barrett--Rudolph threshold, while amplitude damping recovers the exact value of the corresponding exclusion game. In arbitrary dimension, a kernel vector with Schmidt coefficients \(c\) reduces the problem to a universal correlation-matrix optimization whose threshold is \(\lceil \ln 2/\ln(1+c_{\min}/c_{\max})\rceil\). Hence every channel whose Choi kernel contains a full-Schmidt-rank vector becomes perfectly excludable after finitely many uses, with an exact formula when the kernel is one-dimensional. For qubit channels, finite-copy perfect exclusion is equivalent to divergence of the regularized Doeblin information and therefore to an infinite asymptotic retrocausal capacity, attained with zero error after finitely many uses.