Attractors, entropy, and the arrow of time in data space

When data exhibit a preferred direction in state space, the question naturally arises:

Is this direction arbitrary — or does it encode irreversibility?

In physics, directionality without microscopic asymmetry is not unusual. It appears whenever systems are described at a coarse-grained, macroscopic level.

The language for this is non-equilibrium thermodynamics.


Entropy as a geometric concept

Entropy is often introduced as a scalar quantity. But in non-equilibrium systems, it plays a deeper role.

It defines a direction in state space.

When a system relaxes, it does not merely change its state — it moves along directions that increase entropy production.

This motion does not require a force in the Newtonian sense.

It requires only:


Onsager’s insight

Onsager’s reciprocal relations formalize this idea.

Near equilibrium, the evolution of macroscopic variables $X_i$ can be written as:

\[\dot{X}_i = \sum_j L_{ij} \, \frac{\partial S}{\partial X_j}\]

where:

Several crucial points follow immediately:

  1. No force term is introduced
  2. The dynamics are driven by entropy gradients
  3. Time asymmetry enters through response, not through fundamental laws

The arrow of time is not imposed — it emerges.


Attractors as entropy maxima

In this framework, an attractor corresponds to a stationary entropy configuration.

Trajectories converge not because they are “pulled”, but because deviations are dissipated.

The attractor is where

\[\nabla S = 0, \quad \text{but} \quad \delta S < 0 \quad \text{for deviations}\]

Stability is an entropic property.


From physical systems to data space

Now consider an abstract state space reconstructed from observational data.

If:

then the data define an effective arrow of time.

This arrow is not temporal in the coordinate sense — it is organizational.

It tells us which directions are stable and which are not.


Why this is not a new force

The Onsager form shows why introducing a new force is unnecessary.

Directionality arises from:

Calling this a “force” is a matter of language, not physics.

Much like:

the effect is real — but emergent.


Cosmological perspective

In late-time cosmology, background expansion data may be interpreted as trajectories in a macroscopic state space.

If these trajectories:

then cosmic acceleration can be described as entropic relaxation, not as a fundamental interaction.

ΛCDM appears naturally as a stationary state, not necessarily as a source-driven solution.


A final caution

Applying Onsager relations here is not a claim about microscopic gravity.

It is a statement about structure at the level of description.

The argument is diagnostic, not dynamical.

But diagnostics are often where irreversibility first reveals itself.


Closing remark

When data select a direction, and dissipation stabilizes it, an arrow of time has appeared — even if the underlying equations remain time-symmetric.

That arrow lives not in spacetime, but in state space.


Next: Ricci flows, entropic relaxation, and topology in state space
Previous: When data prefers a direction in state space
All Notes : Conceptual and Interpretational Notes