Importance of Nuclear Quantum Effects for NMR Crystallography

Edgar A. Engel, Venkat Kapil, Michele CeriottiThe Journal of Physical Chemistry Letters 2021 · DOI · arXiv

Abstract

The resolving power of solid-state nuclear magnetic resonance (NMR) crystallography depends heavily on the accuracy of computational predictions of NMR chemical shieldings of candidate structures, which are usually taken to be local minima in the potential energy. To test the limits of this approximation, we systematically study the importance of finite-temperature and quantum nuclear fluctuations for <sup>1</sup>H, <sup>13</sup>C, and <sup>15</sup>N shieldings in polymorphs of three paradigmatic molecular crystals: benzene, glycine, and succinic acid. The effect of quantum fluctuations is comparable to the typical errors of shielding predictions for static nuclei with respect to experiments, and their inclusion improves the agreement with measurements, translating to more reliable assignment of the NMR spectra to the correct candidate structure. The use of integrated machine-learning models, trained on first-principles energies and shieldings, renders rigorous sampling of nuclear fluctuations affordable, setting a new standard for the calculations underlying NMR structure determinations.

Introduction

NMR crystallography identifies the 3D structure of materials by comparing experimental chemical shifts with first-principles computational predictions. Traditionally, these predictions rely on "static" calculations, where researchers identify the local minimum of the potential energy surface—the "frozen" geometry—and calculate the NMR shielding at that point.

However, atoms are never truly stationary. Even at absolute zero, they undergo zero-point fluctuations, and thermal motion adds further complexity at higher temperatures. In molecular crystals, such as those found in pharmaceuticals, these motions are often highly anharmonic, rendering simple harmonic vibration models inadequate. The discrepancy between static predictions and experimental reality is often as large as the difference between distinct crystal polymorphs, leading to ambiguous or incorrect structural assignments.

Methodology

To address these limitations, this study employs an integrated modeling workflow that leverages machine learning to make rigorous quantum simulations computationally feasible.

Integrated modeling workflow showing the transition from static structures to (PI)MD simulations and the resulting shielding distributions.

The workflow consists of several key components:

  1. Path Integral Molecular Dynamics (PIMD): This method captures nuclear quantum effects (NQEs) by treating nuclei as "beads" on a ring polymer, simulating their quantum-mechanical spread.
  2. Machine Learning Potentials (MLP): To overcome the prohibitive cost of PIMD using density functional theory (DFT), a Behler-Parrinello neural network was trained to mimic the potential energy surface, accelerating sampling by orders of magnitude.
  3. Machine Learning for Shieldings: Separate ML models (ShiftML) were trained to predict NMR chemical shieldings ($^1$H, $^{13}$C, and $^{15}$N) based on local atomic environments, using data from the GIPAW-DFT method.

The study focuses on three paradigmatic molecular crystals: benzene (Forms I and II), succinic acid ($\alpha$ and $\beta$ polymorphs), and glycine ($\alpha, \beta,$ and $\gamma$ polymorphs). These were chosen for their diverse chemical interactions, including van der Waals forces and hydrogen bonding.

Results and Discussion

A primary finding is that NQEs do more than just "blur" NMR signals; they significantly shift the mean shielding values—by up to 1–2 ppm. This shift is substantial enough to alter the identification of the most likely crystal structure.

The research demonstrates that NMR shielding is inherently non-linear. Calculating shieldings on an "average" structure derived from dynamic simulations yields large errors. Instead, one must average the shieldings over the fluctuations themselves. Furthermore, standard harmonic approximations failed to reproduce PIMD results, underscoring the critical role of anharmonicity in molecular flexibility.

Comparison of polymorph assignment probabilities between static and PI NVT methods, demonstrating improved resolving power.

The inclusion of NQEs consistently reduced the discrepancy between theory and experiment. For instance, in glycine, static $^1$H shieldings struggled to distinguish between $\beta$ and $\gamma$ phases. Incorporating NQEs via PIMD increased the Bayesian assignment probability for the correct structure to nearly 80%. The ML models achieved high precision, with Root-Mean-Square Errors (RMSE) of 0.15 ppm for $^1$H, 0.9 ppm for $^{13}$C, and 1.3 ppm for $^{15}$N.

Conclusion

This work establishes that nuclear quantum effects are essential for accurate NMR crystallography. While the current study focuses on relatively simple paradigmatic cases, it provides a robust proof-of-concept for the necessity of dynamic and quantum treatments. Future work will focus on the transferability of these ML models to more complex systems, such as large pharmaceutical molecules with hundreds of atoms per unit cell, where the initial investment in training data remains a significant hurdle.

Summary updated

Abstract

The resolving power of solid-state nuclear magnetic resonance (NMR) crystallography depends heavily on the accuracy of computational predictions of NMR chemical shieldings of candidate structures, which are usually taken to be local minima in the potential energy. To test the limits of this approximation, we systematically study the importance of finite-temperature and quantum nuclear fluctuations for <sup>1</sup>H, <sup>13</sup>C, and <sup>15</sup>N shieldings in polymorphs of three paradigmatic molecular crystals: benzene, glycine, and succinic acid. The effect of quantum fluctuations is comparable to the typical errors of shielding predictions for static nuclei with respect to experiments, and their inclusion improves the agreement with measurements, translating to more reliable assignment of the NMR spectra to the correct candidate structure. The use of integrated machine-learning models, trained on first-principles energies and shieldings, renders rigorous sampling of nuclear fluctuations affordable, setting a new standard for the calculations underlying NMR structure determinations.
BibTeX
@article{engel2021importance,
  author = {Edgar A. Engel and Venkat Kapil and Michele Ceriotti},
  title = {{Importance of Nuclear Quantum Effects for NMR Crystallography}},
  journal = {The Journal of Physical Chemistry Letters},
  year = {2021},
  doi = {10.1021/acs.jpclett.1c01987},
}
Figure