Benchmarking the acceleration of materials discovery by sequential learning

Rohr, Brian A.; Stein, Helge S.; Guevarra, Dan; Wang, Yu; Haber, Joel A.; Aykol, Muratahan; Suram, Santosh K.; Gregoire, John M.

doi:10.1039/c9sc05999g

Cited by 115 publications

(107 citation statements)

References 36 publications

(51 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Es ist aber genau diese unvoreinge-nommene Regression mit quantifizierbarer Unsicherheit, die für Optimierungsprobleme entscheidend ist. Sei es, einen optimalen Spielzug bei Schach und Go zu planen 14) oder das nächste Experiment zur Synthese eines Katalysators: 15) Bei mehr als vier bis sechs Parametern, die variiert werden, sind Computer-Algorithmen fast unschlagbar schnell. 15) Der entscheidende Schritt ist hierbei, die einzelnen, durch Automatisierung, Parallelisierung oder durch maschinelles Lernen beschleunigten Arbeitsabläufe zu verknüpfen.…”

Section: Lichtabsorber-und Katalysatorforschungunclassified

“…Sei es, einen optimalen Spielzug bei Schach und Go zu planen 14) oder das nächste Experiment zur Synthese eines Katalysators: 15) Bei mehr als vier bis sechs Parametern, die variiert werden, sind Computer-Algorithmen fast unschlagbar schnell. 15) Der entscheidende Schritt ist hierbei, die einzelnen, durch Automatisierung, Parallelisierung oder durch maschinelles Lernen beschleunigten Arbeitsabläufe zu verknüpfen. Denn in der Innovationskette eines neuen Materials ist die Maximalgeschwindigkeit stets durch den langsamsten Ablauf oder den langsamsten Transfer bestimmt -ganz, wie wir es aus der Reaktionskinetik kennen.…”

Section: Lichtabsorber-und Katalysatorforschungunclassified

See 1 more Smart Citation

Die Materialsynthesemaschine

Stein¹,

Suta²,

George³

2020

Nachr Chem

View full text Add to dashboard Cite

show abstract

Section: Lichtabsorber-und Katalysatorforschungunclassified

Die Materialsynthesemaschine

Stein¹,

Suta²,

George³

2020

Nachr Chem

View full text Add to dashboard Cite

show abstract

“…3 ) with random substitution at the A-site. Special quasi-random structure (SQS) method, 4 as implemented in the ATAT package, 5,6 , is used to obtain structures for various A-site alloy compositions. Two different supercell sizes 96 and 144-atoms are used and for each A-site composition with multiple (two or three) structures, varying in the orientation of the MA and FA molecules, are considered.…”

Section: Gibbs Free Energy Of Mixing Calculations (Dft)mentioning

confidence: 99%

“…4 This challenge is similar to those faced by other materials communities, including heterogeneous catalysts, alloyed battery electrodes, and high-entropy metal alloys for structural and magnetic materials. [5][6][7] The halide perovskite field and several others require new tools to experimentally navigate these vast spaces efficiently to locate optima and to extract generalisable scientific insights. [8][9][10][11][12][13][14] Machine-learning-based sequential learning approaches (e.g.…”

Section: Introductionmentioning

confidence: 99%

A Physical Data Fusion Approach to Optimize Compositional Stability of Halide Perovskites

Sun

Tiihonen²,

Oviedo³

et al. 2020

Preprint

View full text Add to dashboard Cite

Compositional search within multinary perovskites employing brute force synthesis are prohibitively expensive in large chemical spaces. To identify the most stable multi-cation lead iodide perovskites containing Cs, formamidinium (FA) and methylammonium (MA), we fuse results from density functional theory (DFT) calculations and in situ thin-film degradation test within an end-to-end machine learning (ML) algorithm to inform the compositional optimization of CsxMAyFA1-x-yPbI3. We integrate phase thermodynamics modelling as a probabilistic constraint in a Bayesian optimization (BO) loop, which effectively guides the experimental search while considering both structural and environmental stability. After three optimization rounds and only sampling 1.8% of the compositional space, we identify thin-film compositions centred at Cs0.17MA0.03FA0.80PbI3 that achieve a 3x delay in macroscopic degradation onset under elevated temperature, humidity, and light compared with the more complex state-of-the-art Cs0.05(MA0.17FA0.83)0.95Pb(I0.83Br0.17)3. We find up to 8% of MA can be incorporated into the perovskite structure before stability is significantly compromised. Cs is beneficial at low concentrations, however, beyond 17% is found to contribute to reduced stability. Synchrotron-based grazing-incidence wide-angle X-ray scattering (GIWAXS) further validates that the interplay of chemical decomposition and phase separation governs the non-linear instability landscape of this compositional space. We reveal the detrimental role of the ẟ-CsPbI3 minority phase in accelerating degradation and it can be kinetically suppressed by co-optimising Cs and MA content, providing insights into simplifying perovskite compositions for further environmental stability enhancement. Our approach realizes the effectiveness of ML-enabled data fusion in achieving a holistic, efficient, and physics-informed experimentation for multinary systems, potentially generalisable to materials search in the vast structural and alloyed spaces beyond halide perovskites.

show abstract

“…The results of new experiments or computations are fed back to the model and used in subsequent iterations. Recently demonstrated applications of SL in materials science include systems that learn how to perform only the most valuable or relevant DFT simulations using previous iterations [5,36,23], improve force fields more rapidly for molecular dynamics simulations [22,37,38], or even synthesize carbon nanotubes at new conditions that promote higher yields and higher qualities of product [25]. Multi-fidelity models, on the other hand, combine measurements or experiments performed at varying degrees of accuracy in their fitting and prediction procedures.…”

Section: Introductionmentioning

confidence: 99%

Multi-fidelity Sequential Learning for Accelerated Materials Discovery

Palizhati¹,

Aykol²,

Suram³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

We introduce a new agent-based framework for materials discovery that combines multi-fidelity modeling and sequential learning to lower the number of expensive data acquisitions while maximizing discovery. We demonstrate the framework's capability by simulating a materials discovery campaign using experimental and DFT band gap data. Using these simulations, we determine how different machine learning models and acquisition strategies influence the overall rate of discovery of materials per experiment. The framework demonstrates that including lower fidelity (DFT) data, whether as a-priori knowledge or using in-tandem acquisition, increases the discovery rate of materials suitable for solar photoabsorption. We also show that the performance of a given agent depends on data size, model selection, and acquisition strategy. As such, our framework provides a tool that enables materials scientists to test various acquisition and model hyperparameters to maximize the discovery rate of their own multi-fidelity sequential learning campaigns for materials discovery.

show abstract

Benchmarking the acceleration of materials discovery by sequential learning

Abstract: Benchmarking metrics for materials discovery via sequential learning are presented, to assess the efficacy of existing algorithms and to be scientific in our assessment of accelerated science.

Cited by 115 publications

References 36 publications

Die Materialsynthesemaschine

Die Materialsynthesemaschine

A Physical Data Fusion Approach to Optimize Compositional Stability of Halide Perovskites

Multi-fidelity Sequential Learning for Accelerated Materials Discovery

Contact Info

Product

Resources

About