Recovering full crystallographic orientation in specimen coordinates from polarised μ-FTIR spectra

Technical Summary: Recovering Full Crystallographic Orientation from Polarised μ-FTIR Spectra

The Problem
Fourier-transform infrared (FTIR) spectroscopy is the standard method for quantifying trace hydroxyl (OH) in nominally anhydrous minerals (NAMs) due to its accessibility and speed. However, birefringent crystals absorb infrared light anisotropically, meaning the measured absorbance depends on the orientation of the crystal relative to the polarisation vector. While polarised FTIR can theoretically resolve both the concentration and orientation of OH-bearing defects, this requires prior knowledge of the crystal's orientation relative to the instrument.

Current practices face significant limitations:

1. Unpolarised FTIR: Most studies use unpolarised spectra on randomly oriented crystals. This approach discards orientation information and systematically underestimates OH concentrations unless hundreds of grains are measured, which is often impractical.
2. Pre-orientation: Measuring polarised spectra on pre-oriented crystals requires aligning the sample before measurement, a barrier for fine-grained materials.
3. EBSD Coupling: Pairing FTIR with Electron Backscatter Diffraction (EBSD) solves the orientation problem but requires two instruments, introduces alignment errors (typically ±8°) when transferring samples, and averages orientation data rather than keeping it internally consistent with the spectrum.
4. Existing Spectral Methods: Previous attempts to recover orientation directly from spectra (e.g., Asimow et al., 2006) have limitations: they do not place the crystal in the specimen reference frame (making comparison with other techniques difficult), rely on fixed spectral ranges that may suffer from detector non-linearity, and lack generality for other polarised techniques.

Methodology
The authors present a method to recover the full crystallographic orientation of an anisotropic crystal in specimen coordinates, expressed as Euler angles, using polarised micro-FTIR (μ-FTIR) spectra. The core innovation is the "wavelength-section method," which reconstructs a two-dimensional section of the crystal's anisotropic response.

• Theoretical Framework: The method builds on the transmittance equation (Asimow et al., 2006), which relates transmittance to the principal axes of the crystal, sample thickness, and the orientation of the polarisation vector. Unlike previous approaches that solve for the polarisation vector relative to the crystal, this method solves for the crystal's orientation relative to a fixed external (microscope) frame.
• Reference Frames: The authors define three coupled reference frames: the sample frame, the microscope frame (where the polariser is fixed along the x-axis), and the crystal frame (defined by the unit cell). The orientation is determined by finding the rotation (using Bunge $zxz$ Euler angles) that aligns the measured transmittance data with a theoretical "transmittance envelope" derived from reference spectra along the three principal axes.
• Data Acquisition: The method requires polarised spectra collected over a range of polariser angles ($\alpha$) at a single diagnostic wavelength. Each $(T, \alpha)$ pair is converted into a Cartesian point on the transmittance envelope.
• Optimisation: The algorithm minimises the misfit between the measured points and the theoretical envelope by iteratively applying trial rotations. To handle the non-convex nature of the orientation cost surface (multiple local minima), the authors tested and recommend using a gradient-based optimiser with multiple random initial guesses or a differential evolution global optimiser.
• Wavelength Selection: The method requires a single wavelength where the transmittance envelope is sufficiently anisotropic (principal transmittances differ) and where the sample thickness keeps transmittance within the detector's linear response range (typically 10–90% T). The authors demonstrate this using the silica overtone region (1500–2100 cm⁻¹) in olivine.

Key Contributions

1. Full Orientation Recovery: The method recovers the complete crystallographic orientation (Euler angles) in specimen coordinates, not just the orientation of the polarisation vector relative to the crystal axes. This allows direct comparison with EBSD and other structural data.
2. Single-Wavelength Flexibility: By operating at a single wavelength, the method avoids the need for broad spectral ranges, allowing users to select wavenumbers that avoid detector non-linearity or baseline issues.
3. Open-Source Implementation: The authors provide FTIRkit, an open-source Python tool implementing the method, relying on standard scientific libraries (NumPy, Pandas, SciPy, Matplotlib).
4. Validation: The method is validated through Monte Carlo simulations and experimental data from olivine crystals of known orientation (measured by EBSD).

Results

• Simulation Performance: In noise-free simulations, the method achieves high accuracy. A gradient-based optimiser with 20 initial guesses recovers orientations within 1° of the true value in >99.75% of cases. The median angular accuracy is approximately 0.004°.
• Noise Sensitivity: When Gaussian noise is introduced, accuracy improves monotonically with the number of measurements (sample size). The method is more sensitive to transmittance noise than to angular ($\alpha$) noise. Reducing transmittance noise significantly improves orientation recovery.
• Experimental Validation (Olivine): The method was tested on olivine grains from a Kilbourne Hole xenolith at two thicknesses (447.5 µm and 244 µm).
  • For the thinner sample, 70% of grains (16/23) were recovered within a 25° disorientation threshold compared to EBSD.
  • For the thicker sample, 50% of grains (15/30) met the threshold.
  • Success depends on the specific combination of wavenumber and sample thickness to ensure the transmittance remains within the linear detector range and the envelope shape is distinct.
  • Failures were often attributed to grains where the measured spectra could not be reproduced by the theoretical model (indicating potential issues with the reference standard, sample inclusions, or thickness variations) rather than a failure of the algorithm itself.

Significance and Claims
The authors position this work as a solution to the "orientation problem" in polarised spectroscopy. They claim that by exploiting the anisotropy of the FTIR signal itself, the method eliminates the need for a second instrument (EBSD) and maintains internal consistency between the spectrum and the orientation.

The paper modestly claims that the approach:

• Applies to any birefringent crystalline phase, provided principal-direction reference spectra and a model relating spectrum to orientation are available.
• Can be extended to other polarised techniques (Raman, UV–Vis, laser-direct IR).
• Offers a more flexible alternative to spectrum-range approaches by allowing the user to avoid degraded wavenumbers.

The authors do not claim the method is perfect for all experimental conditions; they explicitly note that success is contingent on selecting appropriate wavelengths and thicknesses to maintain linear detector response and that experimental challenges (sample movement, epoxy deformation) can affect the accuracy of real-world comparisons with EBSD. The primary contribution is the demonstration that full orientation can be recovered from polarised spectra alone, removing a major barrier to routine polarised μ-FTIR analysis.