Since Newton's time, the scientific method has centered on finding a single equation for a single phenomenon. Even modern symbolic regression, which automatically derives formulas from data, silently inherits this assumption: it seeks one optimal curve. But real systems, from rubber to galaxies, switch between mechanisms as conditions change. Alongside this work on the archive are studies where neural networks learn to derive laws of nature or build gravitational maps, yet all of them also assume a single law. This paper takes a step sideways: instead of asking which equation is best, it asks how many exist simultaneously.
The authors developed a four-step pipeline. First, three parallel symbolic regression searches (global, condition-based, and spatial-block-based) collect a pool of equations with different topologies. Then, for each cell in parameter space, the dominance weight of each equation is computed via Bayesian posterior probability. In the third step, a secondary symbolic regression finds analytical laws governing the evolution of these weights, with a monotonicity constraint applied when needed for irreversible processes. Finally, weights are extrapolated to new conditions, and the prediction is a weighted sum of all equations. The number of mechanisms K* is determined automatically via the BIC information criterion.
In all four tested systems (elastomers, boiling, galaxies, droplet evaporation), BIC selected exactly three independent equations. For elastomers, these are a power law for crystallization hardening, a root law for entropic elasticity, and Hooke's linear law; predictions improved by 72% over gradient boosting. In boiling, the monotonicity constraint added another 20% accuracy. On 3269 measurements from 163 galaxies, the system independently found Newtonian, transitional, and 'constant acceleration' laws, and the 2.2-fold difference in the characteristic acceleration a0 between gas-poor and gas-rich galaxies rules out a universal MOND curve with p < $10^{-34}$. In droplet evaporation, the moment of contact line detachment was automatically captured.
This work differs from neighboring studies in that it does not seek a single best formula but maps an entire ecosystem of mechanisms. Neighboring works use neural networks to derive laws or build gravitational maps, yet all remain within the K=1 paradigm. Here, K* becomes the first measurable quantity, opening an orthogonal direction: not improving equations but describing their coexistence. This turns the debate over the universality of a law from a philosophical question into a computational test.
The topic could evolve toward a dynamic theory of law ecology: not a static map but a description of symbiosis, competition, and succession of equations over time. The operator set could be expanded to include trigonometric and special functions. Applications to turbulence, material fatigue, and stellar evolution would test the approach's universality. Creating an interdisciplinary benchmark for diagnosing the number of mechanisms across fields is also promising.
The method applies to materials mechanics, thermophysics, astrophysics, and microfluidics. It could change how universal laws are tested in biology and climatology.
The next step is to apply the framework to metabolic scaling and climate sensitivity, where debates rage over one versus multiple laws. The operator dictionary also needs expansion, and the K-selection algorithm needs improvement.
In scientific computing, the open question remains how to automatically determine the number of regimes in data without prior assumptions. This work offers a concrete BIC criterion and a continuous weight field instead of hard boundaries, addressing the long-standing problem of smooth transitions between mechanisms.
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The paper introduces a system called Physical Law Ecology, which automatically determines the number of coexisting mechanisms K* from data, constructs a map of their dominance, and derives the laws governing regime shifts. The method was tested on elastomers, boiling, galaxies, and evaporating droplets, where BIC selected three equations and predictions improved by 67–72%.
Where its strength is
A key strength is turning the number of mechanisms from a hidden assumption into a measurable quantity, allowing us to map the coexistence of laws rather than picking just one.
Where it could go next
Worth trying
Why this may matter
If the number of coexisting mechanisms becomes a standard measurable quantity, it could change how we model complex systems, shifting the focus from searching for a single law to describing their ecosystem. This is especially important in problems where conditions vary over a wide range.
Nearby works in our archive
The archive is sparse around this work: the closest review is only 0.56, against a usual 0.64. It looks like few have drilled nearby.
A candle, which we see as a single flame, is actually governed by four different equations: heat conduction in the wax, diffusion combustion in the blue zone, radiation in the bright part, and convection in the tongue above the flame.