Prediction of Molecular Single-Photon Emitters: A Materials-Modelling Approach

TL;DR

The paper addresses the challenge of discovering molecular single-photon emitters within vast chemical space by integrating COD-based SMILES screening, fingerprint-based similarity, and first-principles calculations with a machine-learning–augmented embedding strategy centered on DBT in anthracene. It constructs a pipeline that moves from database search to microscopic observables (emission wavelength, oscillator strength, vibronic coupling, and spin-orbit coupling) to screen and rank candidates. Terrylene is validated as a strong emitter in the same host, while 2000909 and 4127216 emerge as promising new candidates, with DPNP and BDPB offered as validation targets; 4127216 additionally provides a chiral emitter avenue for photonics. The approach demonstrates a computationally affordable route to tailor molecular SPEs for specific tasks and points toward future integration with global exploration and photon-statistics modeling to broaden design capabilities.

Abstract

Interfacing light with quantum systems is an integral part of quantum technology, with the most essential building block being single-photon emitters. Although various platforms exist, each with its individual strengths, molecular emitters boast a unique advantage -- namely the flexibility to tailor their design to fit the requirements of a specific task. However, the characteristics of the vast space of possible molecular configurations are challenging to understand and explore. Here, we present a theoretical and computational framework to initiate exploration of this vast potential by integrating database analysis with microscopic predictions. Using a model system of dibenzoterrylene in an anthracene host as benchmark, our approach identifies promising new candidates, among them a chiral molecular emitter. Future extensions of our approach integrated with machine learning routines hold the promise of ultimately unlocking the full potential of molecular quantum light-matter interfaces.

Figures (12)

• Figure 1: Hunting Emitter-Host pairs:(I) Entries from the Crystallography Open Database (COD) that are availiable for substructure search by SMILES (approximately $2\cdot10^5$ structures) are collected. The SMILES strings are converted to bitvecors of fixed size and dimensionality reduction and clustering is performed on the dataset by using the t-SNE and HDBSCAN algorithms. (II) A suitable, known emitter-host pair is chosen as a reference. We will limit ourselves here to DBT in anthracene (illustrated), which is among the most studied and promising candidates. (III) The Tanimoto Index is used as a metric and potential replacement emitters and hosts are ranked by their similarity to the references. Suitable high-scoring candidates are examined and selected. (IV) Further microscopic analysis, such as DFT and molecular dynamics calculations, are performed for a set of relevant observables. This includes for example emission wavelength, oscillator strength, various metrics for spin-orbit coupling, vibronic coupling, and formation energies. (V) Results are evaluated and new emitter-host pairs are proposed.
• Figure 2: (a)Generation of SMILES strings: Schematic showing the generation of SMILES strings from (1) a molecular structure, which is converted into (2) a graph representing the atomic connectivity, and finally into (3) a linear text-based SMILES representation of the molecule.(b)Global structure of dataset: t-SNE plot showing the distribution of molecular data. Known emitters are indicated by green triangles, and DBT is highlighted as a blue triangle. (c)Similarity Ranking: Logarithmic distribution of the Tanimoto index of the materials in the Crystallography Open Database with respect to DBT. The insets show relevant samples from the distribution. They are, from top to down and left to right: DBT, terrylene, 4127216, 2000909, 1555531 and BDPB, where the numbers are COD IDs. The blue square denotes the reference material, green squares denote a known emitter, although not necessarily in anthracene, and the grey squares denote previously unknown candidates.
• Figure 3: Microscopic Analysis: Figures showing (a) the oscillator strength of emission, (b) the excitation wavelength (absorption), (c) the vibronic coupling entropy (with anthracene as host), (d) the binding energy (with anthracene as host), (e) the intersystem crossing, (f) the reverse intersystem crossing, and the (g) ground state intersystem crossing plotted for emitters against their Tanimoto index. The reference emitter (DBT) and three especially interesting candidates (terrylene, 2000909, 4127216) are highlighted.
• Figure 4: Emitter candidates:(a) The recovery of terrylene serves as validation for our exploitation strategy. (b) Perylene illustrates the relevance of intersystem crossing events. (c) Hexa-peri-hexabenzo[7]helicene (COD ID: 4127216), is a chiral emitter with decent performance at long wavelength, a promising building block for chiral photonics. (d) Tetrabenzo[de,hi, op,st]pentacene (COD ID: 2000909), follow a similar synthesis route as DBT and presents an ideal validation point for future experimental studies. Additional validation candidates: (e) DPNP, and (f) BDPB.
• Figure 5: Gaussian Process Classification: The 10-dimensional space {Tanimoto Index, $f^{\text{abs}}_{\text{osc}}$, $f^{\text{em}}_{\text{osc}}$, $\lambda^{\text{abs}}_{\text{emitter}}$, $\lambda^{\text{em}}_{\text{emitter}}$, SOC, rSOC, GS SOC, $\mathcal{S}^{VC}$, $E_{\text{bind}}$} (Table \ref{['tab:table_emitters_smiles']}) is reduced to the two dominant principal components (PC1 and PC2). An emitter is labeled "good" if it performs better than average in the most relevant categories (left) $f^{\text{em}}_{\text{osc}} > \overline{f}^{\text{em}}_{\text{osc}}$, $\lambda^{\text{abs}}_{\text{emitter}} > \lambda^{\text{abs}}_{\text{anthracene}}$, $\mathcal{S}^{\text{VC}} < \overline{\mathcal{S}^{\text{VC}}}$, and $\text{SOC} < \overline{\text{SOC}}$. Right-hand side, the Spin-Orbit Coupling ($\text{SOC}$) is ignored in the labeling, i.e., we do not include any primer for intersystem crossing.