Bi-Rotation Framework: Materials & Vision

• Bi-Rotation Framework is a dual-concept approach that merges controlled polarization rotation in ferroelectrics with optimized rotation matrices in computer vision.
• It employs advanced computational methods such as the Berry-phase formalism and DFPT-based phonon analysis to quantify and engineer dipole reorientation.
• In computer vision, the method optimizes two independent rotation matrices to resolve pose ambiguities, enhancing accuracy and robustness in real-world scenarios.

The Bi-Rotation Framework refers to two formally distinct paradigms in current research: (1) a materials science approach enabling controlled rotation of the spontaneous polarization vector in layered ferroelectric oxides via isovalent doping at the fluorite-like sublattice; and (2) a geometric computer vision methodology for relative pose estimation, built around the simultaneous optimization of two rotation matrices that reduce pose determination to axis-aligned forms. Both frameworks share the structural motif of leveraging bi-rotational variables to transform and analyze physical or geometric states, though in different application domains.

1. Structural Bi-Rotation in Layered Ferroelectrics

In Aurivillius-phase ferroelectrics, notably Bi$_4$Ti$_3$O$_{12}$ (BiT), the Bi-Rotation Framework provides a systematic set of computational and group-theoretic tools to predict and engineer the reorientation of the polarization vector $\mathbf{P}$ through isovalent doping at the Bi sites in the fluorite-like (FL) sublattice. BiT’s monoclinic B1a1 structure features an alternation of three perovskite-like (PL) blocks (Bi$_2$Ti$_3$O$_{10}$) and one fluorite-like (FL: Bi$_2$O$_2$) block along the $c$-axis, with the spontaneous polarization $_3$0 predominantly aligned along the in-plane $_3$1-direction in the pristine ground state (Co et al., 2018). The framework is intended to enable predictive rotation of $_3$2 toward the out-of-plane $_3$3-direction, in direct response to device integration needs.

2. Theoretical Foundations and Computational Formalism

2.1 Modern Theory of Polarization

The polarization change within the Bi-Rotation Framework is quantified using the Berry-phase formalism for ferroelectrics, which computes $_3$4 by integrating the Berry connection over occupied Bloch functions:

$_3$5

This is decomposed as $_3$6. In undoped BiT, $_3$7 and $_3$8.

2.2 Born Effective Charges and Displacement Analysis

Alternatively, the spontaneous polarization is evaluated using the “small-displacement” formula:

$_3$9

where $_{12}$0 are Born effective charges (BECs) and $_{12}$1 are atom-specific displacements. In BiT, FL Bi cations exhibit larger BECs and off-center displacements (ODCs) than their PL counterparts, identifying them as the key symmetry-sensitive sublattice for producing uncompensated dipole moments along $_{12}$2.

3. Phonon Mode and Layer-Resolved Dipole Decomposition

The Bi-Rotation Framework employs detailed phonon mode analysis (using DFPT at $_{12}$3) to identify low-energy (hard) modes. For BiT, modes at $_{12}$4 THz, which disproportionately displace FL layers in the $_{12}$5-direction, generate dominant layer-resolved dipole moments (LRDM$_{12}$6) from the FL Bi sites. This analysis thus directly links dynamic ionic response to static polarization properties.

$_{12}$7

The uncompensated out-of-plane dipole contributions arise primarily from the FL layer, justifying it as an optimal site for targeted doping.

4. Doping-Driven Polarization Rotation

Upon substitutional doping (6.25 at.\%) of FL-Bi with isovalent elements (P, As, Sb), the Aurivillius structure and net symmetry are perturbed. Lattice and internal atomic positions are re-optimized. The resulting polarization components are summarized as follows (numerical benchmarks detailed refer to (Co et al., 2018), Table VI):

System	$_{12}$8	$_{12}$9	$\mathbf{P}$0	$\mathbf{P}$1	Angle to $\mathbf{P}$2	Band gap (eV)
Pure BiT	0	52.255	7.879	52.847	81.43°	2.173
P‐doped	2.642	47.955	35.031	59.445	53.85°	1.386
As‐doped	2.923	45.806	23.523	51.577	62.82°	1.793
Sb‐doped	5.795	39.738	21.014	45.325	62.13°	2.217

The most substantial reorientation is achieved by P-doping, with $\mathbf{P}$3 enhancement $\mathbf{P}$4 and a $\mathbf{P}$5 rotation of $\mathbf{P}$6 towards $\mathbf{P}$7. The underlying mechanism is the disruption of symmetry by less polarizable dopants, unbalancing the FL dipole moment.

5. Generalized Bi-Rotation Workflow for Layered Materials

The Bi-Rotation Framework is encapsulated in a deterministic workflow:

1. Identify symmetry-sensitive sublattices: Analyze BEC and ODC to localize sites responsible for polarization directionality.
2. Phonon and dynamic analysis: Calculate DFPT modes to isolate hard (non-softening) phonon branches and their associated layer-resolved dipole patterns.
3. Layer-resolved dipole computation: Quantify which layers contribute uncompensated dipoles in the target direction.
4. Dopant selection and structural modeling: Choose isovalent dopants that perturb local symmetry while preserving the overall structure (e.g., group V elements for Bi).
5. Supercell construction and geometry relaxation: Substitute at low concentration (1–10 at.\%), relax structure fully.
6. Polarization recomputation: Use updated BECs and displacements to recalculate $\mathbf{P}$8.
7. Validation: Explicitly extract $\mathbf{P}$9 and the rotation angle $_2$0; compare against device application criteria.
8. Electronic and dynamical stability assessment: Inspect band gap, defect states, and soft mode emergence.

Repeated application enables systematic exploration and optimization of polarization rotation in complex layered ferroelectrics, including n-layer Aurivillius, Ruddlesden–Popper, and Dion–Jacobson families (Co et al., 2018).

6. Bi-Rotation in Geometric Computer Vision

Distinct from the materials context, the birotation framework in computer vision refers to a technique for relative pose estimation between camera systems. Here, two independent rotation matrices $_2$1 are optimized to transform point correspondences from two views such that the essential relative pose can be expressed as a pure translation along one of the principal axes (Zhao et al., 4 May 2025). For $_2$2,

$_2$3

Angle-equality constraints, resulting from projection into image coordinates, generate three geometric metrics $_2$4, each underpinning an energy function $_2$5. These are optimized in parallel on the manifold $_2$6 using robust weighting and Gauss–Newton steps. The basis yielding the minimum energy determines the final relative rotation and translation estimate, with ambiguity resolved by sign-disambiguation and initialization bias.

This birotation method has yielded improved accuracy in multiple standard and real-world datasets compared to canonical approaches, with further robustness to initialization and outlier correspondences (Zhao et al., 4 May 2025).

7. Implications, Limitations, and Extensions

The Bi-Rotation Framework in layered ferroelectrics enables deterministic and material-specific engineering of polarization orientation, directly informing device design requiring non-in-plane polarization. Its rigorous modularity—combining symmetry analysis, lattice dynamics, and first-principles polarization computation—permits generalization to a wide class of layered oxides.

In computer vision, the birotation framework addresses the challenge of non-uniqueness in pose recovery by exploiting axis-aligned rectification; optimization over three basis energies yields higher robustness to initialization and measurement noise. The method’s $_2$7-DoF formulation for a $_2$8-DoF problem introduces discrete ambiguities, handled by sign constraints and initialization schemes.

A plausible implication, given the shared emphasis on pairwise rotational transformation in disparate fields, is the potential for future methodological cross-pollination—for instance, birotational strategies for symmetry breaking or coordinate frame alignment in both materials discovery and geometric inference.

• "Polarization rotation in Bi$_2$9Ti$_3$0O$_3$1 by isovalent doping at the fluorite sublattice" (Co et al., 2018)
• "A Birotation Solution for Relative Pose Problems" (Zhao et al., 4 May 2025)