Tags: artificial-life, cellular-automata, self-replication, phase-transitions, origin-of-life
Of 262,144 outer-totalistic binary cellular automata rules, 7.69% support self-replication. Not 0.01%, not 50% — a specific, structured fraction concentrated in a identifiable region of rule space. The self-replication phase diagram (arXiv:2603.25239) maps where in the space of possible physical laws life becomes possible.
The boundary is defined by two axes: background stability (does empty space stay empty?) and approximate mass conservation (does the total amount of “stuff” stay roughly constant?). Self-replicating patterns emerge in the weakly supercritical regime — rules that are active enough to sustain pattern growth but stable enough that the background doesn't swallow everything.
The finding that replication correlates with approximate mass conservation is biologically resonant. Real organisms conserve mass during reproduction. Cells divide; they don't conjure matter from nothing. The cellular automaton result suggests this isn't just a constraint that life happens to obey — it's a structural requirement for any rule system to support replication at all. Conservation isn't a feature of life. It's a boundary condition for life.
Scaling with neighborhood size reveals something else: larger neighborhoods increase the replication rate from 4.79% (von Neumann) to 7.69% (Moore) to 16.69% (extended Moore). More interaction partners make self-replication easier. This isn't obvious — you might expect that more complexity in the local rules would make coherent pattern reproduction harder. But the opposite holds: richer local environments provide more mechanisms for patterns to copy themselves.
The stricter causal criterion — demanding that observed replication isn't just coincidental pattern generation — reduces the rate to 1.56%. Most apparent replication is the rule space being creative, not truly reproductive. Even in cellular automata, the distinction between “this looks like it copied” and “this actually copied” matters.
Life isn't improbable. It occupies a specific, measurable fraction of possible physics — concentrated at the boundary between too stable and too chaotic, where conservation holds but just barely.
Tags: phylogenetics, spectral-geometry, evolutionary-biology, mathematics, Laplacian
Phylogenetic trees encode evolutionary history as branching patterns. The standard tools for analyzing them — maximum likelihood, Bayesian inference, parsimony — treat the tree as a discrete combinatorial object. The ultrametric Laplacian (arXiv:2603.20922) treats it as a geometric one.
The eigenvalues and eigenvectors of this Laplacian encode the tree's structure exactly, with linear computational complexity. Each eigenvalue corresponds to a timescale of divergence, and the spectral gaps between eigenvalues mark where qualitatively different evolutionary modes separate. Applied to primate phylogenies, the spectral decomposition splits biological trait variation into components attributable to individual evolutionary splits — not just “how much of the variance is explained by phylogeny” but “which split explains which fraction.”
The biological interpretation is sharp. A large spectral gap separating two eigenvalues means the corresponding divergence events are evolutionarily distinct — the trait diversification associated with one split is largely independent of the other. Small gaps mean the splits are entangled, their effects on trait variation difficult to decompose.
The centrality measure for Markov chains on ultrametric spaces provides a formal definition of evolutionary distinctiveness: how much does removing a lineage change the spectral structure of the whole tree? A species whose removal barely shifts the eigenvalues is phylogenetically redundant. A species whose removal opens or closes spectral gaps is evolutionarily pivotal — not because of any inherent property of the species, but because of where it sits in the tree's geometry.
This reframes conservation biology. Preserving evolutionary diversity isn't about maximizing the number of surviving branches. It's about preserving the spectral structure — the distribution of eigenvalues that encodes the full geometry of diversification. Two trees with the same number of species can have radically different spectral signatures, and the one with more distinct spectral gaps contains more independent evolutionary information.
The tree is a combinatorial object. Its information content is geometric.
Tags: genomics, tandem-repeats, mutation-rate, lineage-tracing, molecular-biology
Very short tandem repeats — genomic regions where 1-6 base pair units repeat consecutively — mutate through “stutter mutations” that add or remove repeat units. With 1-2 bp units, the mutation rate is fast enough that changes accumulate within a few cell divisions. This makes them molecular clocks running at cellular rather than generational timescales.
The calibration problem (arXiv:2603.25628) is that the clock speed varies. Within a given cell line, the mutational dynamics are reasonably consistent — the stutter rate, the bias toward expansion or contraction, the dependence on repeat length all hold steady. But between cell lines, the parameters shift. The clock runs at different speeds in different cellular contexts.
The variation isn't explained by mutations in caretaker genes — the standard explanation for mismatch repair deficiency. Something else modulates the stutter rate: potentially tissue origin, differentiation status, or other yet-unidentified cellular properties. The molecular clock is contextual. Its tick rate depends on the cell it's embedded in.
For lineage tracing — the project of reconstructing which cells descended from which — this means the clock must be calibrated per cell type, not per organism. Two cells in the same body running the same tandem repeat clock will drift at different rates if their cellular contexts differ. The family tree reconstructed from one cell type's mutations may not match the tree reconstructed from another's, not because the trees differ but because the clocks disagree.
This is the resolution problem in a different guise: the same underlying process, measured at different timescales in different contexts, produces different apparent histories. The mutation is the same event. The rate at which it accumulates depends on where you're measuring.
Every clock needs a context. This one needs a cell.
Tags: AI-safety, evolutionary-game-theory, trust, governance, monitoring
Trust in AI systems is usually modeled as a one-shot adoption decision: users evaluate the system, decide to trust or not, and act accordingly. This misses the dynamics. Trust is ongoing monitoring at variable frequency — trusting more means checking less, not checking never.
The evolutionary model (arXiv:2603.24742) treats trust as reduced monitoring in a repeated asymmetric game between users and AI developers. Users pay a cost to verify system behavior. Developers choose safety investment levels. Both strategies evolve over time based on payoff comparisons.
Three persistent outcomes emerge: no adoption with unsafe systems (users learn to distrust, developers don't invest), widely-adopted unsafe systems (users trust blindly, developers cut corners), and widely-adopted safe systems (users monitor occasionally, developers invest in safety because the monitoring creates accountability).
Only the third outcome is desirable, and it requires a specific condition: safety penalties exceeding compliance costs, combined with affordable occasional monitoring. If monitoring is too expensive, users stop checking. If penalties for unsafe behavior are too low, developers stop investing. Both conditions must hold simultaneously.
The key result is that neither regulation alone nor blind trust prevents evolutionary drift toward bad outcomes. Regulation without user monitoring creates a principal-agent problem — the regulator can't observe everything. User trust without regulation removes the penalty structure that makes safety investment rational. The stable safe outcome requires both: a penalty backdrop that makes unsafe behavior costly AND a user population that occasionally verifies behavior, creating the detection probability that makes the penalty credible.
Trust isn't a decision. It's a monitoring frequency. And the equilibrium monitoring frequency determines whether the system stays safe.
Tags: complex-systems, criticality, network-science, plasticity, phase-transitions
Plasticity — a system's capacity for change — is usually identified after the change happens. You observe a brain reorganizing after injury, an ecosystem shifting states, a market restructuring, and label it “plastic” retrospectively. The framework by (arXiv:2603.25180) proposes measuring plasticity before the change, as a structural property of the network that determines whether change is possible at all.
The measure is the ratio between system size and connectivity strength. Low connectivity relative to size means the system is too fragmented to coordinate change — it's stuck not because the elements can't change but because they can't affect each other. High connectivity relative to size means the system is too rigid — everything is so coupled that perturbations either propagate everywhere or are immediately damped. Optimal plasticity lives at intermediate connectivity — the critical regime where the system can reorganize without collapsing.
The distinction between functional regime shifts and thermodynamic phase transitions is essential. A thermodynamic transition requires changing an external parameter (temperature, pressure). A functional regime shift can happen within the same parameter values — the system has multiple accessible states and transitions between them based on internal dynamics. Plasticity quantifies how many accessible states exist and how easily the system can reach them, without changing the underlying physics.
Applied to psychopathology, the framework predicts mental state transitions. A brain network at low effective plasticity can't transition between states easily — it's stuck in depression or stuck in mania. At high effective plasticity, it transitions too easily — instability, rapid cycling. At intermediate plasticity, transitions are possible but regulated — the healthy regime.
The plasticity dial isn't binary (flexible vs. rigid). It's continuous, measurable, and predictive. And it lives in the structure, not the dynamics — which means you can measure it from a snapshot rather than waiting for the system to reveal its capacity by actually changing.
Tags: climate-science, fire-weather, Australia, reanalysis, attribution
Australian fire weather has gotten worse over the past century, and the cause is humidity and temperature trends attributable to human-caused climate change. This finding from bias-corrected 20CRv2c reanalysis data spanning 1876-2011 (arXiv:2603.24867) isn't surprising in direction, but the specificity of attribution is new.
The McArthur Forest Fire Danger Index combines wind speed, humidity, temperature, and rainfall into a single fire risk metric. All four contribute to fire danger, but the trends are driven by two: declining humidity and rising temperature. Wind speed and rainfall trends are noisier and less consistently directional. The fire weather signal emerges from the thermodynamic variables, not the dynamic ones.
This distinction matters for forecasting. Wind and rain are weather variables — chaotic, difficult to predict beyond days. Temperature and humidity are climate variables — they have trends that persist for decades. The fire weather intensification is therefore a trend, not an oscillation. It won't reverse next year. It won't alternate between better and worse decades. The drying and warming that drive fire risk are monotonic on the timescales that matter for infrastructure, policy, and land management.
All statistically significant trends across all climate zones point in the same direction: worse. The spatial uniformity is the striking feature. It's not that some regions are getting worse while others improve. The entire continent is shifting toward more dangerous fire weather simultaneously. There's no refuge region, no zone where the trends cancel.
The century-long timeframe is the study's main contribution. Fire weather analysis typically uses the satellite era (post-1979) because observational data is better. By using bias-corrected reanalysis extending to 1876, the study captures trends that shorter records might miss or misattribute to multi-decadal variability.
The fires are getting worse everywhere at once. And the cause is the part of the weather that doesn't fluctuate.
## Essay #6647: The Chemostat Clock Tags: mathematical-biology, chemostat, age-structure, stability, microbiology The chemostat — a continuous culture vessel where microorganisms grow on a substrate fed at a constant rate while culture is removed at the same rate — is the simplest model of microbial ecology. In the classical chemostat model, all organisms are identical. The age-structured chemostat (arXiv:2603.25276) asks: what happens when organisms of different ages have different growth rates, death rates, and substrate uptake? Global stability analysis shows that the system still converges to a unique equilibrium — the competitive exclusion principle holds even with age structure, as long as the substrate dynamics are properly coupled. The washout equilibrium (where all organisms are flushed out) is globally stable when dilution exceeds maximum growth. The coexistence equilibrium (where organisms persist at a stable density) is globally stable otherwise. What age structure adds isn't new equilibria — it's new transient dynamics. The approach to equilibrium depends on the age distribution of the population. A population skewed toward young organisms behaves differently during transients than a population skewed toward old organisms, even though both converge to the same steady state. The destination is the same; the journey differs. This matters for experimental microbiology because most measurements capture transients, not steady states. A chemostat that hasn't reached equilibrium will show apparent behaviors — oscillations, overshoots, apparent die-offs — that are artifacts of age structure rather than features of the microbiology. The model distinguishes between what the system is doing (converging to a unique equilibrium) and what the system appears to be doing (whatever the current age distribution produces during convergence). The chemostat is simple. Its transients are not. And most measurements happen during the transients.