[Paper Review] In situ Learning-Based Spin Engineering of Pulsed Dynamic Nuclear Polarization

Filip Jensen, José P. Carvalho+7 more|arXiv (Cornell University)|Mar 19, 2026

Advanced NMR Techniques and Applications0 citations

TL;DR

The paper pioneers in situ Bayesian optimization to design pulsed DNP sequences directly on the spectrometer, achieving broadband, high-enhancement polarization transfer for trityl and TEMPO systems, with comparisons to existing methods.

ABSTRACT

Pulsed Dynamic Nuclear Polarization (DNP) is currently receiving substantial interest as a means to enhance the sensitivity of nuclear magnetic resonance (NMR) and magnetic resonance imaging (MRI) by orders of magnitude. It has also received much attention as a central ingredient in many modalities of electron spin-involved quantum sensing. Relative to spin engineering associated with NMR, the design of efficient pulsed DNP experiments with a broad experimental scope are challenged by large electron-nuclear spin systems, large electron spin-involved interactions, and instrumental non-idealities and limitations. All of this may challenge traditional NMR-like theoretical and numerical pulse sequence engineering. Exploiting state-of-the-art instrumentation and taking advantage of the high sensitivity of DNP relative to NMR, we here demonstrate the use of combinations of Bayesian machine learning methods and constrained random walk procedures to design pulse sequences extit{in situ}, by experiments, directly on the spin systems responding to spectrometer instructions. For trityl and nitroxide samples, it is demonstrated that efficient broadband DNP pulse sequences can be designed in situ with experimental protocols benchmarked against in silico analogs.

Motivation & Objective

• Motivate and address the challenge of designing efficient pulsed DNP experiments for large, complex electron-nuclear spin systems.
• Demonstrate in situ Bayesian optimization to design DNP pulse sequences directly from experimental feedback.
• Benchmark in situ learning against existing analytical and numerical design methods using representative spin systems (trityl and TEMPO).
• Show that constrained random walk and Bayesian methods can produce broadband, high-efficiency DNP transfer under realistic instrumental limitations.

Proposed method

• Use Bayesian machine learning to optimize DNP pulse sequences directly from experimental feedback (no gradient information).
• Combine Bayesian optimization with constrained random walk (cRW) to guide search toward resonant transfer conditions (ZQ/DQ).
• Employ 6- and 24-pulse DNP transfer elements and variable repetition to achieve broadband offsets.
• Compare in situ optimized sequences to MONTE CARLO, NOVEL, PLATO, and cRW-OPT baselines.
• Test on trityl OX063 and TEMPO nitroxide samples at 80 K in deuterated glycerol-water systems.

Figure 1 : Schematic representation of the experimental setup used for in situ Bayesian development of DNP pulse sequences. a) The generation of amplitude and phase modulated pulses for an initial flip (or excitation) pulse and a periodic DNP transfer element in combination mediating electron to pro

Experimental results

Research questions

• RQ1 Can in situ Bayesian optimization discover high-efficiency, broadband DNP pulse sequences on real spectrometer hardware?
• RQ2 How does constrained search (cRW) influence the optimization and robustness to MW inhomogeneity?
• RQ3 Do BayesOpt-designed sequences match or exceed performance of established sequences (NOVEL, PLATO, cRW-OPT) on model spin systems?
• RQ4 How well do optimized excitation pulses cooperate with existing DNP transfer elements to maximize polarization transfer?
• RQ5 What is the relative performance of Bayesian optimization for narrow-line (trityl) versus broad-line (TEMPO) samples?

Key findings

• Bayesian optimization rapidly improves DNP transfer, achieving higher enhancements than Monte Carlo for trityl and TEMPO samples.
• For trityl, a 24-pulse BayesOpt sequence yields an enhancement factor approaching 15 and broadband profiles comparable to cRW-OPT1, with standout performance near ZQ/DQ resonances (kI values).
• Constrained Bayesian optimization focusing on ZQ resonance (kI = 2) converges faster to high transfer efficiencies and yields broadband sequences.
• Combined optimized excitation pulse plus DNP transfer element demonstrates superior broadband transfer compared to a single hard excitation pulse in trityl systems.
• In TEMPO, Bayesian optimization outperforms random MC optimization (enhancement factors 13 vs. 9) and can outperform a spin-lock baseline under certain conditions.
• Experimental results show that BayesOpt results align with, but also reveal deviations from, two-spin simulations, highlighting instrument and spin-system complexities captured by in situ learning.

Figure 2 : in situ optimization of the DNP transfer element in Fig. 1 a with 6 (a-c) and 24 (e,f) pulses (modulation time $t_{m}$ =120 ns; 49 MHz maximal MW field strength; initial flip pulse of amplitude 47 MHz and duration 5 ns, see Fig. 1 a). The measurements were performed at 80 K using a trityl