Multi-Variable Multi-Metric Optimization of Self-Assembled Photocatalytic CO2 Reduction Performance Using Machine Learning Algorithms

Abstract

The sunlight-driven reduction of CO₂ into fuels and platform chemicals is a promising approach to enable a circular economy. However, established optimization approaches are poorly suited to multivariable multimetric photocatalytic systems because they aim to optimize one performance metric while sacrificing the others and thereby limit overall system performance. Herein, we address this multimetric challenge by defining a metric for holistic system performance that takes multiple figures of merit into account, and employ a machine learning algorithm to efficiently guide our experiments through the large parameter matrix to make holistic optimization accessible for human experimentalists. As a test platform, we employ a five-component system that self-assembles into photocatalytic micelles for CO₂-to-CO reduction, which we experimentally optimized to simultaneously improve yield, quantum yield, turnover number, and frequency while maintaining high selectivity. Leveraging the data set with machine learning algorithms allows quantification of each parameter's effect on overall system performance. The buffer concentration is unexpectedly revealed as the dominating parameter for optimal photocatalytic activity, and is nearly four times more important than the catalyst concentration. The expanded use and standardization of this methodology to define and optimize holistic performance will accelerate progress in different areas of catalysis by providing unprecedented insights into performance bottlenecks, enhancing comparability, and taking results beyond comparison of subjective figures of merit.

Figure 1

(A) Molecular components of photocatalytic system and reaction scheme; (B) photocatalytic test results with 1.5 μM CoPyP_C16, 30 μM Rubpy_C17 or [Ru(bpy)₃]Cl₂ (Rubpy), including or excluding 225 μM C₁₂E₆ (∼3 CMC), 100 mM NaHAsc, phosphate (0.1 M), CO₂-sat. (pH 6.3) after 15 min 447 nm illumination with 2.3 W LED at 25 °C with 250 rpm orbital shaking, 1 mL reaction volume. Product detection by GC, 3 replicates except Rubpy_C17 + C₁₂E₆ where n = 21 over 8 batches. Tabulated values in Table S1.

Figure 2

(A) Photoluminescence decay showing lifetime of excited photosensitizer in absence of reductant (excitation 460 nm, detection 620 nm); (B) transient absorption spectroscopy (pump 460 nm, probe 510 nm) measuring lifetime of reductively quenched photosensitizer in the presence of reductant; and (C) transient absorption spectroscopy (pump 460 nm) measuring spectral changes after 10 and 1000 μs delay. Thirty μM photosensitizer in phosphate buffer (PB, 0.1 M, pH 7.0), Ar purged. Five μM CoPyP_C16, 225 μM C₁₂E₆ and/or 100 mM NaHAsc as indicated. Dashed lines show data fitting to determine lifetimes.

Figure 3

Heuristic optimization of photocatalytic system showing catalyst turnover frequency (TOF) and Yield of CO as a function of catalyst (A), photosensitizer (B), surfactant (C), reductant (D) and buffer (E) concentration. Unless specified, 1.5 μM CoPyP_C16, 30 μM Rubpy_C17, 225 μM C₁₂E₆ (∼3 CMC), 100 mM NaHAsc, phosphate (0.1 M), CO₂-sat. (pH 6.3) after 15 min (60 min for catalyst series) 447 nm illumination with 2.3 W LED at 25 °C with 250 rpm orbital shaking, 1 mL volume, GC quantification, 3 replicates.

Figure 4

Holistic optimization using learning algorithms. Overview of the workflow (A); sorted data to show holistic improvement measured by objective function (eq 1 with w₁ = 0.4, w₂ = 0.4, w₃ = 0.2) alongside improvement of both Yield_CO and TOF_CO (B); and sorted data showing ascending Yield_CO and corresponding TOF_CO (C). The data in B and C are sorted and not in chronological order. Full experimental results tabulated in Table S7. Experimental conditions: various concentrations of CoPyP_C16, Rubpy_C17, C₁₂E₆, NaHAsc and phosphate buffer, CO₂-sat. (pH 6.3) under 447 nm illumination with 2.3 W LED at 25 °C with 250 rpm orbital shaking, 1 mL reaction volume for 15 min, GC quantification.

Figure 5

Machine learning analysis showing the normalized importance of each system component to holistic optimization by objective function 1 (A), Yield_CO and QY_CO (B), TON_CO and TOF_CO (C); with machine learning predictions with the Random-forest regression models (inset), where model performance is shown by coefficient of determination R², mean absolute error (MAE), and Root Mean Squared Error (RMSE).