Ready to Work Together?
Accelerated Material Design
Materials Innovation through AI-Quantum Synergy
Materials Innovation through AI-Quantum Synergy
Materials Innovation through AI-Quantum Synergy
Symmetry-Aware Diffusion Models for Crystalline Material Discovery
Crystalline materials represented as infinite arrays of repeating unit cells, with atoms of defined chemical elements positioned at precise lattice points. These materials are distinguished by their inherent symmetries, which play a central role in determining their physical and chemical properties.
Intractābilis has developed a pioneering generative framework that exploits this representation of material symmetries. By combining an innovative neural network design with a discrete diffusion process, the model is able to generate novel protostructures that extend beyond known crystallographic forms. Rather than specifying exact atomic coordinates, these symmetries are captured within protostructure descriptions that are organised according to space groups, ensuring that essential structural information is retained while unnecessary detail is abstracted. This approach captures the structural diversity of crystalline materials while maintaining alignment with fundamental physical constraints.
A selected subset of the generated protostructures is further developed into complete crystal structures. For the most promising candidates, atomic geometries are refined using a pretrained, broadly applicable interatomic potential. This enables accurate prediction of structural stability and material performance, reducing the need for computationally expensive simulations.
Our framework demonstrates strong performance across several key measures, including the diversity of generated materials, the novelty of structural motifs, and consistency with the underlying distribution of crystallographic data. These results underline the potential of our approach to accelerate materials discovery and provide new pathways for advanced applications.
Reinforcement Learning from Human Feedback for Topological Material Discovery
Topological materials represent an expansive class of materials whose electronic behaviour is dictated by the topology of their band structures. Among these, topological insulators and topological crystalline insulators are of particular significance. They are characterised by insulating bulk states and protected surface or edge states that remain robust against impurities, defects, and disorder. Such features make these materials highly promising for applications in quantum computing, spintronics, and energy-efficient electronic devices. Identifying and realising materials with these properties remains one of the most pressing challenges in contemporary materials science.
Intractābilis has developed an innovative solution based on reinforcement learning from human feedback (RLHF) that provides an effective route towards the discovery of novel topological materials. At the foundation of this approach lies a state-of-the-art generative model for materials design, which is first trained on a broad set of structures and subsequently fine-tuned for topological discovery. The RLHF mechanism involves guiding the generative model through the reward model capable of predicting topological properties.
Our methodology ensures that the generative model retains its ability to produce stable and diverse material candidates while significantly improving the likelihood of generating systems with non-trivial topological characteristics. Our solution advances the search for materials capable of enabling the next generation of quantum and electronic technologies.
Variational Quantum Algorithms for Multiconfigurational Aromatic Materials
Aromatic molecules serve as fundamental components in many organic materials employed across optical and electronic technologies, including light-emitting diodes, solar cells and semiconductors. A comprehensive understanding of both their ground and excited states is essential for advancing the development of innovative functional materials. In particular, the electronic states of aromatic molecules with π-conjugated systems often display pronounced multiconfigurational characteristics, making them challenging to model with conventional approaches.
Intractābilis has pioneered a novel variational quantum algorithm applied initially to the naphthalene molecule and enabled the identification of conditions that determine which electron configurations make the most substantial contributions to the electronic states of interest. Our method achieves accuracy that exceeds the capabilities of classical techniques. Our framework has shown that excitation energies can be determined with quantitative precision. Extending the approach to tetracene has further demonstrated the versatility and robustness of the methodology.
As quantum computing hardware continues to advance, the potential to harness larger numbers of qubits will expand. This progression will gradually increase the proportion of calculations that can be carried out through quantum rather than classical computation, opening the way to highly scalable and more efficient solutions for electronic structure prediction.
By bringing together structural modelling, domain-specific retrieval, autonomous research agents, and multi-objective optimisation, Intractābilis is creating a foundation for AI and quantum augmented materials science that is both scalable and interpretable. Our solutions offer industrial research teams a decisive edge in the race to discover, optimise, and deploy next-generation materials—reducing cost, compressing timelines, and uncovering insights previously hidden in complexity.
© 2025 Intractābilis