Resolve the confounding between polarization state and domain-wall proximity

Determine whether the effects of polarization state and domain-wall proximity on polarization switching dynamics can be distinguished independently when the accessible experimental field of view lacks admissible locations combining low polarization-state and low wall-proximity scores.

Background

The autonomous PFM experiment was intended to investigate how local domain structure affects polarization switching dynamics. However, measurements targeting low-polarization-state regions were repeatedly close to domain walls, creating a confounding relationship between polarization state and wall proximity.

The complete admissible field of view contained no pixel with both polarization-state and wall-proximity scores below 0.33, and the same combination remained absent under less restrictive thresholds. Consequently, additional spectroscopy within the same frame could not provide the independent comparison needed to resolve the respective effects of these two variables. Resolving this issue requires an experimental state space or measurement design that supplies independently varying polarization-state and wall-proximity conditions.

References

The trajectory also revealed an unresolved scientific contrast (Figure 9C). Measurements selected to probe low-polarization-state regions were repeatedly also close to domain walls. The agent recognized that polarization state and wall proximity were therefore not being sampled independently and returned to the low-polarization criterion at decisions 7, 8, and 12 in an attempt to obtain a low-polarization, low-wall-proximity comparison. Analysis of the complete admissible field of view shows why these attempts could not succeed. No admissible pixel had both polarization-state and wall-proximity scores below 0.33, and the same combination remained absent when both thresholds were increased to 0.40 or 0.50. Among pixels with polarization-state score below 0.33, wall proximity had a minimum of 0.53 and a median of 0.94. The requested comparison therefore did not exist within the available experimental state space (Figure 9C).

— Hierarchical automation of scanning probe microscopy through agentic orchestration and algorithmic control  (2609.04015 - Slautin et al., 3 Sep 2026) in Section 3.5, “Point-selection audit, scientific inference, and experimental identifiability”