🔬 optics

Predicting THz Generation Capability of Organic Crystals through Data Mining and Crystal Nonlinearity Models

This paper demonstrates that combining DFT computations with mathematical models to predict nonlinear susceptibility from crystal structures effectively identifies organic crystals with THz generation capabilities comparable to or exceeding state-of-the-art materials, while establishing a comprehensive tensor-based approach as the standard for evaluating nonlinear optical properties.

Original authors: Sin Hang Ho, Ashton Roma, Connor Barlow, Matthew Lutz, Natalie Green, Stacey Smith, David Michaelis, Jeremy A Johnson

Published 2026-08-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sin Hang Ho, Ashton Roma, Connor Barlow, Matthew Lutz, Natalie Green, Stacey Smith, David Michaelis, Jeremy A Johnson

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Light is more than just what we see; it is a wave of energy that can be coaxed into behaving in surprising ways. When a powerful beam of laser light strikes certain solid materials, the material does not simply absorb the energy or let it pass through unchanged. Instead, the atoms within the material rearrange their internal electrical charges in a complex, rhythmic response. This reaction creates new colors of light that were not present in the original beam, or it can generate invisible pulses of energy that sit between the frequencies of radio waves and infrared light. These invisible pulses are known as terahertz radiation, a type of energy with immense potential for medical imaging, security scanning, and high-speed communication. The key to unlocking this potential lies in finding the right solid crystals to act as the engine for this conversion. For decades, scientists have relied on a few specific materials to perform this task, but the search for better, more efficient crystals has been a slow and difficult process of trial and error.

A team of researchers at Brigham Young University has taken a different approach, bypassing the slow grind of physical experimentation to look directly at the blueprint of matter itself. They turned to a massive digital library called the Cambridge Structural Database, which contains the precise atomic maps of over 1.2 million known crystal structures. From this vast collection, they isolated nearly 78,000 crystals that lack a center of symmetry, a specific geometric requirement that allows a material to generate these new frequencies of light. Using powerful computer simulations, the team calculated how each of these crystals would respond to a laser beam. They did not just guess; they modeled the behavior of the individual molecules inside the crystal and then calculated how those molecules would work together when packed into a solid block. This allowed them to predict, with high precision, which crystals would produce the strongest electrical response and, consequently, the most powerful terahertz pulses.

The results of this digital mining operation were striking. The researchers found that many crystals, which were originally discovered for other purposes or had never been considered for this specific job, showed a similar or larger ability to generate terahertz light than the current industry standards. In fact, several of the newly identified candidates produced predicted signals significantly stronger than the best-known materials currently in use. One of the most promising new candidates, a crystal with the code name ASEKUH, showed a predicted performance that far exceeded the capabilities of the famous DAST crystal, which has long been a top choice for scientists. Another candidate, LIGXEG, also demonstrated exceptional potential. These findings suggest that the world of terahertz generation is far richer than previously thought, hiding powerful materials in plain sight within existing chemical records.

To ensure their computer models were not just producing theoretical fantasies, the team cross-checked their predictions against real-world experiments. They compared their calculated values for known, high-performing crystals with actual measurements taken in laboratories. The match was remarkably close. When the computer predicted a crystal would be strong, the lab measurements confirmed it was strong. When the model predicted a specific orientation of the crystal would work best, the experiments agreed. This validation gave the researchers confidence that their method could be trusted to identify new materials before a single atom was grown in a lab. It also highlighted a critical flaw in how scientists have traditionally estimated these properties. For years, many researchers have relied on simplified mathematical shortcuts to guess a crystal's performance, assuming that the internal structure of the material behaves in a uniform, predictable way.

The study rigorously tested these shortcuts and found them to be dangerously unreliable. When the researchers compared the results of the simplified formulas against their detailed, comprehensive computer models, they discovered that the shortcuts often failed to capture the true complexity of the crystal's internal geometry. In many cases, the simplified methods overestimated the performance of a crystal by a wide margin, or they missed the mark entirely because they ignored the specific angles at which the molecules were packed. The study showed that these shortcuts only worked for a small fraction of the crystals and failed to account for the subtle, directional nature of how light interacts with matter. This means that relying on old, simplified rules could lead scientists to waste time and resources chasing materials that are actually quite weak, while overlooking the truly powerful ones that require a more careful analysis to discover.

The implications of this work extend beyond just finding a better crystal for a single device. By proving that a comprehensive, detailed approach is necessary to accurately predict how a material will behave, the researchers have established a new standard for the field. They demonstrated that the intricate dance of molecules within a crystal cannot be reduced to a simple average; every angle and every interaction matters. This insight allows scientists to screen thousands of potential materials in a fraction of the time it would take to grow and test them physically. The study does not claim to have solved the problem of terahertz generation, nor does it promise that these new crystals are ready for immediate commercial use. However, it provides a clear, verified map of where the best materials are hiding. It suggests that the next generation of high-performance optical devices may not come from inventing entirely new chemicals, but from recognizing the hidden potential in crystals that have already been sitting on a shelf for years, waiting to be understood.

Technical Summary: Predicting THz Generation Capability of Organic Crystals through Data Mining and Crystal Nonlinearity Models

Problem Statement
The development of efficient nonlinear optical (NLO) materials for frequency conversion, particularly for generating high-intensity broadband terahertz (THz) radiation, relies heavily on identifying crystals with large second-order nonlinear susceptibility coefficients ( $\chi^{(2)}$ ). Experimentally determining these properties is resource-intensive, requiring the growth of high-quality single crystals in multiple orientations to measure all tensor elements. Furthermore, existing computational frameworks for predicting NLO performance often rely on simplified approximations that may not accurately capture the complex directional dependencies of molecular hyperpolarizability within specific crystal packings. There is a critical need for a robust, high-throughput computational method to screen vast databases of crystal structures for potential THz generators and to validate whether simplified models are sufficient for such predictions.

Methodology
The authors employed a combined data mining and computational workflow to evaluate the NLO potential of organic crystals:

Data Mining and Curation: From the Cambridge Structural Database (CSD), containing over 1.2 million entries, the authors extracted 77,843 non-centrosymmetric crystal structures. This dataset was expanded from previous work to include ionic structures and co-crystals, not just single-component neutral molecules. After filtering for structural disorders (e.g., overlapping atomic positions), a final dataset of 76,175 candidate structures remained.
DFT Calculations: Molecular hyperpolarizability tensors ( $\beta_{ijk}$ ) were computed for the individual molecular components (ions or chromophores) of each structure using Density Functional Theory (DFT). To avoid spurious inter-component charge transfer artifacts, calculations were performed on isolated molecular entities rather than the entire unit cell.
Macroscopic Modeling: The microscopic $\beta_{ijk}$ values were integrated with crystallographic data (space group, unit cell dimensions, density) using a comprehensive mathematical model (Eqs. 3–4) to calculate the macroscopic second-order nonlinear susceptibility tensor ( $\chi^{(2)}_{IJK}$ ) and the $d_{IJK}$ tensor. Local field factors were approximated using the refractive index of the well-known crystal DAST due to the lack of specific data for the vast majority of mined structures.
Polarization Calculation: The induced nonlinear dielectric polarization ( $P_{NL}$ ) was calculated for the three primary crystal faces under various incident light polarizations. The maximum induced polarization ( $P_{max}$ ) served as the primary metric for ranking the THz generation potential of each crystal.
Validation and Comparison: The computational framework was validated by comparing calculated $d_{IJK}$ values and $P_{max}$ trends against experimental THz generation efficiencies for known crystals (DAST, OH1, PNPA, etc.). Additionally, the authors rigorously compared their "complete model" (Eqs. 3–4) against commonly used simplified approximations (Eqs. 7–8) that treat hyperpolarizability as a scalar or use single-angle projections.

Key Results

Validation of the Model: The computational results demonstrated strong consistency with experimental data. For known THz generators like DAST and OH1, the calculated $d_{IJK}$ values matched experimental measurements within experimental error margins. Furthermore, the ranking of calculated $P_{max}$ values (specifically $(P_{max})^2$ ) correlated directly with experimentally measured THz conversion efficiencies for crystals such as PNPA, DAST, and OH1.
Discovery of New Candidates: The screening identified several structures with $P_{max}$ values exceeding those of state-of-the-art THz generators. Notably, the ionic crystal ASEKUH showed a calculated $P_{max}$ of 208 mC/m² (compared to 54.9 mC/m² for DAST), and LIGXEG showed 136 mC/m². Other candidates like XAFTEE, CUDSAX, and XEGMEB also exhibited superior or comparable predicted performance.
Limitations of Simplified Approximations: A critical finding was the significant discrepancy between the complete model and simplified approximations (specifically Eq. 7, which uses a scalar $\beta_{max}$ $β_{ma x}$ and a crystal packing order parameter).
- Only 56.4% of the diagonal $d_{III}$ values calculated via the simplified model fell within $\pm 50\%$ of the complete model's values.
- For 381 structures, the simplified model overestimated values by a factor of 5 or more.
- The study identified that the accuracy of simplified models depends heavily on the angle between the polar axis and the crystallographic axis ( $\theta_P$ ) and the relative magnitude of diagonal versus off-diagonal tensor elements. When $\theta_P > 45^\circ$ or when off-diagonal elements are significant, the simplified model fails to provide accurate predictions.

Significance and Claims
The paper claims that the proposed workflow—combining DFT with a comprehensive tensor projection model—provides a reliable method for predicting the THz generation capability of organic crystals prior to synthesis. This approach allows for the efficient screening of large databases to identify high-potential NLO candidates, thereby directing experimental resources toward the most promising materials.

Crucially, the authors assert that the "complete model" approach should be the standard method for evaluating the nonlinear optical properties of single-crystalline materials. They demonstrate that commonly used simplifications, while occasionally accurate for specific symmetric systems, are insufficient for general screening as they often overlook critical directional dependencies and off-diagonal tensor contributions, leading to significant errors in predicting material performance. The study highlights that many existing crystal structures, originally designed for other purposes, possess latent potential for high-field THz generation that can only be accurately revealed through this rigorous computational framework.

More like this