Research scope & scientific positioning

Neomundi Lab conducts research at the intersection of thermodynamics, information theory, and computation, with a specific focus on artificial intelligence and large-scale computational systems.

Modern AI systems are increasingly constrained not by algorithmic ingenuity alone, but by energetic limits, irreversibility, and long-term stability at scale. Despite this, most contemporary approaches treat computation as an abstract informational process, largely decoupled from its physical and thermodynamic substrate.

Law E introduces a different scientific stance.

It is a thermodynamic–informational framework designed to formalize the coupling between energy, information, and computation, and to study computational systems as physical systems governed by energetic constraints.

Rather than proposing new learning paradigms or cognitive models, this research focuses on the conditions under which complex computational systems remain coherent, efficient, and stable over time.

To our knowledge, Law E constitutes the first operational framework explicitly applying thermodynamic–informational principles to artificial intelligence and computational infrastructures.

Observed phenomena

Experimental and simulation work conducted within the Law E framework has revealed a set of robust, recurrent phenomena, observed independently of architectures, optimization strategies, or learning mechanisms.

These observations include, but are not limited to:

• Emergent energetic sobriety

Under explicit energetic constraints, certain computational trajectories consistently dissipate less energy than others, without optimization, intelligence, or coordination.

Sobriety emerges as a structural property of the system rather than a designed objective.

• Structural symmetries

Stable regimes exhibit recurrent informational and energetic symmetries that persist across scales and initialization conditions, suggesting invariant organizational patterns.

• Threshold and phase-transition effects

Beyond specific energetic or informational thresholds, systems undergo irreversible transitions toward instability, runaway dissipation, or loss of coherence.

These phenomena are treated as empirical observations, not assumptions. Their interpretation and formalization remain active research topics.

Experimental and simulation context (high-level)

The research does not aim to simulate intelligence, cognition, or real-world datacenters directly.

Instead, minimal experimental systems are constructed to isolate thermodynamic effects in discrete computational processes, under controlled energetic and informational constraints.

Key characteristics of the experimental approach include:

• no learning or optimization objectives

• no reward functions

• no intelligence assumptions

• explicit energetic accounting

• discrete, observable state transitions

This methodology enables the identification of structural properties expected to generalize to real-world AI systems and large-scale computational infrastructures.

Detailed experimental protocols, simulations, and results are documented separately in dedicated research publications.

Open research interfaces

To support reproducibility, exploration, and collaboration, Law E is made accessible through multiple research-oriented interfaces, available upon request.

These include:

• API-level exploration

Programmatic access to thermodynamic metrics, constraints, and governance primitives derived from Law E.

• Constraint-based UI exploration

Interactive interfaces enabling researchers to explore system behavior under explicit energetic and informational constraints.

• Datacenter homeostasis dashboards

Observational tools designed to study stability, dissipation, and long-term coherence in large-scale computational infrastructures.

These interfaces are intended for scientific investigation and applied research, not as consumer-facing products.

From framework to operating system

Law E is not limited to a conceptual framework.

It is currently being implemented as a thermodynamic operating layer, composed of minimal governance primitives designed to regulate how computation unfolds over time.

This operating layer does not replace models or architectures.

It provides constraints, measurements, and regulation mechanisms governing computational dynamics.

As of today, three core primitives have been implemented and validated in experimental contexts. Additional primitives are under active development.

Publications and ongoing work

Formal developments, experimental results, and applied proofs of concept are documented in dedicated research papers and technical notes.

Ongoing work includes:

>> https://doi.org/10.5281/zenodo.18483571
> Information–Energy dynamics and endogenous regulation in Artificial Systems (Loi E)

> Filtre E — audit trace example information-energy regulation.pdf

> Information-energy dynamics simulator-Loi E — Experimental observations.pdf

> Measured effects of endogenous regulation in large language models.pdf

• A complete experimental paper detailing the Law E simulation results

• Proofs of concept for governance mechanisms (Reg–Sel, Filtre E)

• Applied studies targeting AI systems and large-scale computational infrastructures

>> https://doi.org/10.5281/zenodo.17861148

> Law E: An Operational Thermodynamic Framework for Autonomous Cognitive Systems.pdf

Foundational paper presenting the Law E framework for information–energy dynamics and autonomous regulation in cognitive systems.

Further publications will be released as the research progresses.