Skip to content

Interaction-Parameter Sweep

Experimental method — instantiates Controlled Demixing and Domain Formation

Varies the interaction-controlling knobs across a grid to find where separation switches on, how sharp the threshold is, and how wide the safe operating window runs.

An Interaction-Parameter Sweep is the systematic experiment that maps how a mixture's tendency to separate responds to the variables you can actually control. Rather than asking "does this separate?" at one condition, it steps a controlling knob — temperature, composition ratio, salt, solvent quality, molecular weight — across a range and records, at each setting, whether and how the system demixes. Its defining output is a response surface: the threshold where separation begins, how steeply behavior changes near it (a knife-edge or a gentle slope), the width of the robust window where you get the intended outcome, and — as conditions deepen — which pathway the separation takes. It is the tool that turns a black box into a dialed map of cause and effect over the controllable conditions.

Example

A formulator developing a complex coacervate — a dense, polymer-rich droplet phase used to encapsulate a fragrance — needs to know where, in the space of ingredients, the two-phase window actually lives and how forgiving it is. They run an interaction-parameter sweep over two knobs at once: the mixing ratio of the two oppositely-charged polymers, and the salt (ionic strength) that screens their attraction. A plate of dozens of wells, each a different ratio–salt pair, is prepared and read for turbidity and phase volume.

The resulting grid is a map. It shows a coacervation window centered near charge balance that narrows and then vanishes as salt rises past a threshold (screening kills the attraction) — and it shows the edges: a knife-edge on the high-salt side where a tiny excess collapses the phase, and a broad, forgiving plateau near the center. That map is the deliverable. It tells the formulator not just a working recipe but how much margin each knob has, which is what lets the product survive real-world variation in dosing and water hardness. Deepening the sweep also reveals where droplets give way to a spontaneous bicontinuous texture — a change of pathway, not just of degree.

How it works

The method treats the controllable variables as axes and populates a designed grid of conditions, measuring a separation readout (onset, phase fraction, domain character) at each point. From the grid it extracts three things a single experiment cannot: the location of thresholds, the sensitivity (local slope) around them, and the width of robust windows.[1] Pushed across the range, it also classifies pathway — sorting conditions into no-separation, nucleation-and-growth, and spontaneous/bicontinuous regimes — because the same knobs that trigger separation also select how it proceeds. It is exploratory and offline: it characterizes the landscape, it does not run the process.

Tuning parameters

  • Which variables are swept — the choice of axes. Sweep the knob that actually drives the interaction and the map is predictive; sweep an inert one and it is noise.
  • Range and resolution — how wide and how finely each axis is stepped. Coarse steps find the window cheaply but can jump over a knife-edge; fine steps resolve it at higher cost.
  • Dimensionality — one variable at a time versus a full factorial grid. One-at-a-time is cheap but blind to interactions between knobs; multi-factor grids catch coupling but explode in size.
  • Readout sensitivity — what counts as "separated" (turbidity, phase fraction, a morphology signal). A blunt readout misses weak or early separation; a sharp one is costlier per point.
  • Replication near edges — extra sampling where behavior changes fast, to pin a threshold whose exact location matters, at the expense of breadth elsewhere.

When it helps, and when it misleads

Its strength is converting guesswork into a map with margins: it locates thresholds, exposes how sharp or forgiving they are, and defines the robust operating window — the difference between a recipe that works once and one that survives real variation. It is also the natural way to discover pathway boundaries, where separation changes kind rather than just degree.

It misleads mainly through under-sampling a sharp landscape: a coarse grid can straddle a knife-edge and report a smooth, safe window that does not exist, or miss a narrow window entirely.[1] A one-variable-at-a-time sweep is blind to interactions — two knobs each "safe" alone can be hostile together — and any sweep is only as valid as its readout's ability to see weak separation. The classic misuse is running a thin sweep, finding one working point, and reporting it as the condition without ever mapping the margin around it. The discipline is to match grid resolution to how sharply behavior changes, sweep the coupled variables together where interaction is plausible, and report the window and its edges, not a single lucky point.

How it implements the components

  • controllable_condition_set — it operationalizes the set of controllable variables, establishing which knobs actually move the separation and over what range each is effective.
  • demixing_pathway_classifier — by reading behavior across the swept range it sorts conditions into regimes (no separation, nucleation-and-growth, spontaneous/bicontinuous), classifying how the system demixes, not just whether.

It does not record the curated, qualitative verdicts on which pairings are allowed — that is Compatibility Matrix — and it does not track the live margin to the boundary during an actual run; that is Phase-Boundary Monitor. The sweep maps the response landscape offline; it neither judges nor monitors.

Notes

A sweep and a phase diagram are close cousins pointed at different questions: the sweep asks how sensitively the outcome responds to the knobs you can turn (margins and pathways over controllable variables), while phase-diagram mapping asks where the equilibrium boundaries sit in thermodynamic state space. A sweep can be run without ever committing to a full diagram, and is the faster route when operating margin, not equilibrium structure, is the pressing unknown.

References

[1] In polymer mixtures the drive to separate is captured by the Flory–Huggins interaction parameter, which typically scales inversely with temperature and combines with composition and chain length to set whether a blend is one phase or two. A sweep is, in effect, an empirical traverse of that parameter's influence — which is why the sharpness of the miscibility threshold, and the risk of stepping over it between grid points, is a real experimental hazard. ↩