Why connected state evolution matters.

The previous post in this series argued that emergence and enumeration ask different questions. This one takes the distinction one layer deeper. Even within an emergent study, there is a further choice: whether the output is a population of microstates the system can reach, or a connected sequence of microstates the system actually moves through.

Both are honest answers. They are not the same answer. The difference between them is the difference between a snapshot library and a film.

What a population misses.

Suppose an emergent study returns a thousand microstates the system is capable of occupying across an operating range. Each microstate is fully characterized. Bitstrings, spin observables, correlations, the works. A researcher can browse them, compute statistics on them, ask which ones share what features.

What that population cannot say is which microstates the system moves between, and in what order. The cell is at one configuration. Then it is at another. The question of which configuration follows which is not answered by knowing the population. It is answered by the trajectory.

Path-dependent behavior — hysteresis, memory effects, state-of-health drift — lives in the trajectory, not in the population. Two cells with identical microstate populations can behave differently if they traverse those microstates in different orders. The population is necessary; it is not sufficient.

What connected evolution provides.

A connected state evolution is a sequence of emergent microstates in which each adjacent pair is validated against physical continuity criteria. The state at frame N is not just an arbitrary member of the population; it is the state the system reaches starting from frame N−1 under the engine's evolution. Frame N+1 follows N for the same reason. The sequence is a path.

Built correctly, a connected evolution covers a full cycle — charge or discharge in the battery case — from one end of the operating range to the other. Every microstate along the way is the validated successor of the one before it.

The product of this construction is what we call a master seam. The construction itself is a non-trivial pipeline, because the physical continuity criteria are strict and the engine produces more candidate microstates than will fit on any single path. Most of them belong on adjacent paths. The work is figuring out which microstates connect to which.

What a master seam shows.

A population can tell you the cell reaches certain configurations. A master seam tells you the order. That ordering is where the answers to the cell-level questions actually live.

Consider charge-discharge hysteresis. The cell's voltage at a given state of charge during charging is not the same as its voltage at the same state of charge during discharging. The population of microstates the cell occupies at that state of charge cannot, by itself, explain the hysteresis. The explanation is in which microstates the cell reaches first on one direction versus the other — the trajectory carries the asymmetry that the population averages away.

The same logic applies to degradation. A cell that has cycled a thousand times has visited a great many microstates. Knowing which microstates is useful. Knowing which microstates it visited in which order — which states the system tended to dwell in, which transitions accelerated late in life, which parts of the trajectory shifted as the cell aged — is where the failure mechanisms become readable.

A study that returns isolated microstates can describe what happened. A study that returns connected trajectories can describe how it happened. For a research program trying to predict and prevent future failures, the second description is the actionable one.

The validation is the work.

A trajectory that has not been validated against physical continuity criteria is just a list in an arbitrary order. The work of building a master seam is almost entirely the work of checking that each adjacent transition is one the system would actually make. Bitstrings, spin observables, transverse coherence, and basis-mismatch signatures of two candidate adjacent states must satisfy tolerance bands derived from the seam's own local variability. Candidates that fail the criteria are rejected. The sequence is reassembled until every adjacent pair clears the checks.

What this gets right is the local continuity: each step from frame to frame is a step the engine's evolution actually produces. What it does not assume is that every transition along a real cycle is gentle. Some transitions are. Others are not.

What we are still learning to read.

Across the master seams produced so far, there are points where the playback does not look smooth. A handful of qubits shift or flip in a way that registers as a visible jump rather than a gradient. These are not random. They occur at consistent states of charge across independent seams, on the same material, in the same direction of cycling. Reproducibility argues against them being artifacts.

What they are, exactly, is something we are working on. They may be branch points, where the trajectory splits between distinct quantum configurations and the seam captures one branch crossing. They may be local phase boundaries, where the system reorganizes discontinuously in microstate space without changing macroscopic coordinates much. They may be something else again. The interpretation is open. What is closed is that they appear reproducibly, and that the validation criteria accept them as physically continuous in the engine's evolution.

From what the magnet team has shared, they are looking at something analogous in their TFIM sweep: a reproducible coupling window where the ordered program's structure changes qualitatively over post-circuit lattice evolution, with the transition occurring at consistent parameter values across independent runs. Different system, different coordinate, but the same shape of pattern — structural change that reproduces, surfaces from the trajectory rather than the snapshot, and resists easy categorization until more runs come in. Their forthcoming release through benchmarks.iqintel.io will document what they have found on their own terms.

The honest position is that smooth playback is not the test for whether a seam is correct. Validation against the continuity criteria is the test. Smooth playback is the common case; reproducible discontinuities are part of what the engine is showing us. Treating those discontinuities as failures would discard real structure. Treating them uncritically would overclaim. We sit with the question and keep watching.

Why most simulation tools do not produce this.

Most simulation methods produce configurations at the parameters the user supplied, with no notion of continuity between them. Two adjacent points in a parameter sweep may be physically discontinuous — they may sit in different basins of attraction, on different sides of a phase boundary, or simply be unrelated by any evolution the system would actually undergo. The method does not know and does not check.

For some research questions this does not matter. If the goal is to map out which configurations are energetically favorable, the lack of trajectory information is not a flaw. The methodology is fit for the question.

For questions about how a cell behaves in operation — which is most of the questions a battery program actually has — the absence of trajectory information is exactly the gap that prevents the simulation output from being directly useful. The output describes possible states. It does not describe what the cell does.

Subatomic Computing was built to close that gap. The engine emerges microstates; the seam construction validates and connects them; the output is what the cell does, frame by frame, across its operating range. The trajectory is the deliverable. The configurations are the building blocks.

The film, not the snapshot library.

A snapshot library is useful. It is not what a researcher investigating path-dependent behavior actually needs. The researcher needs the film — the connected, validated, frame-by-frame record of how the system moves from one state to the next, including the parts that move smoothly and the parts that do not. Subatomic Computing produces that film. The fact that it can be played in a viewer is incidental. The fact that it is built against physical continuity criteria is the deliverable.

The final post in this series will turn to a specific finding that surfaced when we ran this methodology on a real cathode system. The finding is the kind of result that connected evolution makes legible and that a snapshot library would have averaged into silence.