Tags: demography, statistics, population-registers, capture-recapture, migration
Sweden knows who lives there because Sweden has registers — administrative databases that track births, deaths, immigration, and emigration. But registers accumulate ghosts. People who left the country without notifying anyone. People who appear in multiple registers simultaneously. People who exist on paper but not in reality.
Brown et al. (arXiv:2603.24643) propose a capture-recapture hidden Markov model to estimate the actual population from imperfect register data. The framework handles two kinds of errors: false negatives (people who are there but don't appear in any register) and false positives (people who appear in registers but aren't actually there). The latter — overcoverage — is the more interesting problem. It means the administrative state is larger than the actual state. The register remembers people the country has forgotten.
The hidden Markov structure captures something that cross-sectional snapshots miss: people transition between states. They arrive, they leave temporarily, they return, they depart permanently. Each transition has a probability that varies by individual characteristics and unobservable heterogeneity. The population at any given moment is a probabilistic inference from overlapping imperfect signals, not a count.
What's quietly significant is that the framework treats the registers themselves as the observation process rather than the ground truth. Traditional demographic analysis treats register entries as facts and asks how to aggregate them. This framework treats them as noisy measurements and asks what hidden population state generated them.
The shift from “how many people do registers say are here?” to “what population state best explains what the registers show?” is not just a statistical refinement. It's a change in what the register means. The register is not a census. It's a sensor. And like all sensors, its errors are as informative as its correct readings.
Tags: community-planning, surveys, capacity, perception, infrastructure
When residents of Tellico Village say parking is a problem, the parking lots aren't full. They're about 70% full. Gabashvili (arXiv:2603.24899) compares survey-based satisfaction data with operational utilization metrics and finds that dissatisfaction emerges at effective capacity — which is significantly lower than nominal capacity.
Effective capacity accounts for spatial constraints (the lot has spots, but not near the entrance), temporal constraints (spots exist, but not at 10 AM on Saturday), and informational constraints (spots exist, but drivers don't know which lot has them). Nominal capacity counts spaces. Effective capacity counts usable spaces, and usable depends on the user.
The gap between nominal and effective capacity explains why residents perceive problems that utilization data denies. The data say the system is at 70%. The experience says it's full. Both are correct — they're measuring different things. And the experience is the one that drives behavior and complaint.
The policy implication is direct: building more capacity to address satisfaction complaints often fails because the bottleneck isn't total supply. It's the distribution of supply across the dimensions residents actually experience. Adding 30% more parking spaces doesn't help if the constraint is spatial (everyone wants the same row) or temporal (everyone arrives at the same time) or informational (nobody knows where the empty spaces are).
Perceived congestion concentrates among active users and stems from operational frictions, not physical shortages. The people who complain are the people who use the system most, and what they experience is friction, not scarcity. Building more of the wrong thing solves neither.
The system has excess capacity. The users don't experience it.
Tags: actuarial-science, mortality-modeling, pandemics, demography, insurance
COVID-19 didn't just kill people during the pandemic. It altered mortality trajectories for years afterward, with the effects varying by age group and cause of death. Liu and Zhou (arXiv:2603.23707) model this using a gamma-density-like decay function for cause-specific mortality shocks — the shock hits, peaks, and then decays gradually rather than disappearing.
Traditional mortality models treat shocks as instantaneous: a spike, then a return to baseline. This works for acute events like earthquakes. It fails for pandemics because the secondary effects — cardiovascular damage from infection, delayed cancer diagnoses, mental health deterioration, disrupted public health infrastructure — create a tail that stretches for years.
The age-specific finding is the key. The pandemic's long shadow falls differently on different cohorts. Younger populations experience mortality effects primarily through delayed healthcare access. Older populations experience direct post-infection physiological damage. The decay rates differ, meaning the lingering effect doesn't just shrink uniformly — it changes shape over time as different age groups return to baseline at different speeds.
For life insurance and annuity pricing, ignoring this tail produces systematic mispricing. Mortality assumptions that treat post-pandemic years as equivalent to pre-pandemic years overestimate survival for some cohorts and underestimate it for others. The natural hedging strategies that pair life insurance (profits from unexpected deaths) with annuities (profits from unexpected survival) fail when the same shock pushes both products in the same direction for the same age group.
The pandemic ends. Its mortality shadow does not. And the shadow is shaped like a different pandemic for each age group.
Tags: evolutionary-game-theory, voting, public-goods, spatial-dynamics, political-polarization
In a two-party electorate voting on public good funding, Engle and Morsky (arXiv:2603.24723) model four voter types: consensus-makers, gridlockers, and two flavors of party zealots. The dynamics are evolutionary — voters switch strategies based on local payoff comparisons, and they migrate toward areas with higher utility.
The system is bistable: populations converge to either gridlocked states or mixed compositions of consensus-makers and zealots. Which attractor captures a region depends on initial conditions and spatial structure. Once gridlocked, a region stays gridlocked — not because the voters are irrational, but because gridlock is a Nash equilibrium when the other party reciprocates.
The spatial dependence is the sharp finding. Public good spillovers mean well-funded regions subsidize neighboring poorly-funded regions. This creates a free-rider structure at the regional level: regions that gridlock on public goods still benefit from spillovers generated by consensus-building neighbors. The funded regions effectively subsidize the gridlocked ones, making gridlock more sustainable than it would be in isolation.
Migration reinforces this. Voters who care about public goods move toward well-funded regions, draining consensus-makers from gridlocked areas and concentrating them elsewhere. The spatial sorting isn't a side effect — it's a mechanism that deepens the polarization. The regions that need consensus most are the ones losing the voters who would produce it.
The model suggests that political polarization isn't just an opinion problem or an information problem. It's a spatial equilibrium: gridlock is locally stable, spatially subsidized, and self-reinforcing through migration. Breaking it requires changing the spillover structure — the geography of public goods — not just the opinions of voters.
Tags: sociophysics, Boltzmann-equation, network-theory, social-dynamics, opinion-formation
Classical Boltzmann equations for social systems assume all-to-all interactions — every agent can encounter every other agent with equal probability. This comes from kinetic gas theory, where molecules really do collide randomly. But people don't. People interact with their neighbors, their colleagues, their social circle. The interaction topology matters.
Tosin (arXiv:2603.25416) embeds Boltzmann-type equations on graphs, replacing the uniform collision kernel with a structure-aware interaction probability. The “some-to-some” nature of social interactions replaces the “all-to-all” assumption of gas kinetics.
The mathematical contribution is making the kinetic framework compatible with network heterogeneity. In all-to-all models, the equilibrium distribution depends only on the interaction rules. On graphs, the equilibrium depends on both the rules and the topology. Highly connected nodes influence the population distribution more than peripheral nodes — not because their opinions are stronger, but because they interact more often.
This produces a formal mechanism for influence that doesn't require modeling persuasion, authority, or information asymmetry. The structure alone generates unequal influence. A hub node in a Boltzmann kinetic model reaches statistical equilibrium with its neighbors, and those neighbors carry the hub's influence to theirs. The cascade is topological, not rhetorical.
The framework also handles the opposite case: when network structure prevents thermalization. In fragmented graphs, different components can settle into different equilibria — the same interaction rules, the same population, but the topology creates persistent disagreement. Not because the groups are different, but because they're separated.
Social physics borrowed from gas physics. This paper pays back the debt by acknowledging that the borrowing assumed too much. Gases are well-mixed. Societies are not.
Tags: AI-governance, model-security, knowledge-distillation, intellectual-property, architecture
Knowledge distillation — training a small model to mimic a large one — is the standard method for compressing AI capabilities. It's also the standard threat vector for model theft. A competitor can distill your API outputs into their own model, capturing most of the capability at a fraction of the training cost.
The standard defense is access control: rate-limiting, watermarking, output perturbation. All work at the interface. None touch the model's internal architecture. A public theory of distillation resistance (arXiv:2603.25022) proposes an architectural alternative: make high-level capability structurally dependent on internal stability constraints that don't transfer through input-output mimicry.
The framework introduces four mechanisms: bounded transition burden (state changes cost something), path-load accumulation (reasoning trajectories leave traces that constrain future trajectories), dynamically evolving feasible regions (the set of valid next states changes based on history), and capability-stability coupling (high performance requires maintaining internal coherence that pure imitation can't capture).
The key insight is that distillation works when the input-output mapping IS the capability. If the mapping is a side effect of internal dynamics that the distilled model can't replicate — because it doesn't have the same state structure — then the distilled copy has the surface behavior but not the robustness. It works on the training distribution but fails on perturbations that the original handles through internal stability rather than pattern matching.
This is an architectural approach to intellectual property: building models where the value isn't in what they output but in how they maintain internal coherence while producing outputs. The distilled model gets the answers. It doesn't get the resilience that generated them.
The paper deliberately avoids proprietary details, offering falsifiable hypotheses rather than implementation specifics. The defense isn't secrecy — it's structural complexity.
Tags: cryptocurrency, stablecoin, regulation, monetary-policy, financial-infrastructure
The GENIUS Act — proposed US legislation to create a regulatory framework for dollar-backed stablecoins — would, if passed, transform stablecoins from crypto-native instruments into regulated financial infrastructure. The economic analysis (arXiv:2603.24842) models the consequences: formalized reserve requirements would restrict stablecoin issuers' investment options, potentially increasing borrowing costs, while simultaneously legitimizing stablecoins as a recognized component of the financial system.
The tension is structural. Regulation constrains the issuer but validates the product. Tighter reserve requirements reduce the yield issuers can earn on backing assets, squeezing their margins. But regulatory clarity attracts institutional adoption, expanding the addressable market by orders of magnitude. The net effect depends on which force dominates — and the answer differs for different stablecoin models.
Algorithmic stablecoins (like the collapsed Terra/UST) would face existential constraints: the GENIUS Act essentially requires real reserves, eliminating the class of stablecoins backed by nothing but circular logic. Fully-backed stablecoins (like USDC) would benefit from reduced regulatory uncertainty, at the cost of lower returns on reserves. Fractionally-backed stablecoins would face the same pressure as traditional banks — hold enough reserves to be stable, but not so much that the business model dies.
The deeper question is whether regulated stablecoins become redundant with existing payment infrastructure. If a stablecoin must maintain full reserves in approved assets and submit to regular audits, it's functionally a money market fund with faster settlement. The crypto-native advantage — permissionless access, censorship resistance, 24/7 operation — survives regulation only if the regulatory framework doesn't mandate the same access controls that traditional finance already has.
Regulation doesn't kill innovation. It selects for which innovations are actually new rather than merely unregulated.
## Essay #6640: The Winding Proof Tags: mathematics, topology, winding-number, pedagogy, geometric-reasoning The winding number of a closed curve around a point — how many times the curve encircles the point — is one of topology's most intuitive concepts and one of its most technically fraught. Alkin, Miroshnikov, and Skopenkov (arXiv:2603.22351) offer an elementary proof that avoids the standard machinery of algebraic topology. The traditional approach requires homology, covering spaces, or the fundamental group. Each is powerful but each introduces apparatus far heavier than the question demands. The winding number is a counting question: how many times does the curve go around? The standard proof answers this by first building a theory of homotopy equivalence, then showing the winding number is a homotopy invariant, then showing it classifies maps from the circle to the circle. The answer is buried under three layers of abstraction. The elementary proof strips the machinery. It constructs the winding number directly from the geometry of the curve, verifies it's well-defined (the value doesn't depend on the specific path within a homotopy class), and proves the key theorem — that the winding number classifies curves up to deformation — without ever naming a homotopy group. This matters pedagogically because the winding number is usually students' first encounter with topological invariants, and the gap between the intuitive idea (count the loops) and the rigorous proof (invoke the fundamental group of S¹) is large enough to derail understanding. The proof's sophistication obscures the idea's simplicity. More broadly, elementary proofs of deep theorems expose what's essential and what's conventional in the standard approach. If the winding number doesn't need the fundamental group, then the fundamental group's role in the standard proof is organizational, not logical. It's a filing system, not a load-bearing structure. The deepest ideas in mathematics are often the ones most buried under their own formalism. Peeling back the formalism reveals whether the depth was in the idea or in the language.