absorb.md

Russell Kaplan

Chronological feed of everything captured from Russell Kaplan.

ldmppr: An R Package That Drops the Mark-Location Independence Assumption in Spatial Point Processes

The `ldmppr` R package introduces a practical framework for modeling location-dependent marked spatial point processes, explicitly relaxing the widely-used but often unrealistic assumption that marks are independent of point locations. The package supports the full modeling workflow — estimation, simulation, goodness-of-fit evaluation, and visualization — making it directly applicable to real-world datasets, particularly in forestry. This fills a methodological gap where existing tools default to mark-location independence, which breaks down in many applied settings.

Commutativity of Generic Derivations as a Necessary and Sufficient Condition for Supersimplicity and Superrosiness in Geometric Fields

This paper establishes sharp equivalences between model-theoretic complexity classes for geometric theories of fields: stability collapses to strong minimality, simplicity to SU-rank 1, and rosiness to surgicality. A key corollary, combining with Hrushovski's prior work, is that algebraically bounded stable fields are exactly expansions of algebraically closed fields by constants. For fields equipped with generic tuples of derivations, the paper proves that both supersimplicity (in the algebraically bounded simple case) and superrosiness (in the o-minimal case) are equivalent to the derivations commuting, with explicit rank bounds via the Kolchin polynomial.

Asteroid Family Population Size Drives Collisional Resurfacing, Counteracting Space Weathering

A study of 154 asteroid families using NEOWISE thermal infrared photometry finds that family population size — not age — is the dominant predictor of fresh surface signatures. The newly introduced V-Dominance Index (VDI), measuring the fraction of family members with p_V/p_IR > 1.2, correlates significantly with population size across both silicaceous and carbonaceous taxonomic types. Critically, large ancient families (>2 Gyr) still exhibit statistically significant fresh surface tails, indicating that elevated collisional activity in massive families continuously counteracts space weathering saturation. This reframes resurfacing as a population-driven, ongoing process rather than a purely age-dependent one.

Diffusion Models for Robot Motion Planning Gain Robustness via Dynamic Sum-of-Costs Collision Guidance

This paper introduces a diffusion-based motion planning method for robotic manipulation that steers the denoising process using the gradient of a summed collision cost across all obstacles — rather than individual or static cost terms. A key contribution is a dynamic scheme for selecting the step at which gradient guidance is injected, which improves stability and generalization. On the MPiNets benchmark, the method outperforms competing classical and deep learning approaches across diverse test configurations. The work directly addresses the well-known generalization gap that plagues existing learned planners when encountering out-of-distribution environments.

Twisted BiSb Moiré Superlattices Enable Electrically Reconfigurable Topological Domains via Structural Relaxation

Structural relaxation in twisted bilayer BiSb spontaneously generates coexisting topologically nontrivial (Z₂=1) and trivial (Z₂=0) domains within a single moiré unit cell, a phenomenon that deepens with decreasing twist angle. The internal domain boundaries host topologically-protected gapless 1D edge states observable via simulated STM. Critically, both the domain structure and associated edge states can be reversibly tuned by an out-of-plane electric field, establishing a route to programmable, electrically reconfigurable topological circuitry in moiré van der Waals systems.