†University of California,

Using Molecular Dynamics Simulations to Detect Reaction Johnson *

Liu ,
Davis
Marshall Hutchings†,
Events: A Case Study with Iron Carbonyl Clusters Lee-Ping Wang† *University of the Pacific

Motivation for Developing Binary Bond Order Time Series from

Sample Fe3(CO)9 Isomers
New Analysis Tools Molecular Dynamics Simulations
• A molecule can have many theoretical isomers, each existing • Molecular dynamics simulations were run using the • Molecular dynamics simulations are used to explore the • The optimized structures are clustered together into discrete
within local minima throughout the molecule’s potential energy BP86 and B3LYP functionals with a modified DZP potential energy surface of the iron carbonyl compounds. groups of isomers using a bond order-distance metric. The
surface. Dunning basis set. • The trajectory of a high-temperature molecular dynamics clusters produce a standard set of reaction events in the
• The computational discovery of these isomers requires the ability • Theoretical isomers of several iron carbonyls simulation is able to cross energy barriers and enter the wells molecular dynamics trajectory.
to overcome energy barriers that separate local minima. published in 2006 by Wang, King, and Schaefer2 were of multiple energy minima. • While the method of clustering using a bond order-distance
• Molecular dynamics simulations are useful in traversing these • Geometry optimizations are done on each frame in the metric to distinguish between unique isomers is robust, it is
used as the initial geometries.
barriers. molecular dynamics simulations to converge structures into computational costly since every frame in the molecular
• A range of temperatures were used for the molecular
• The transition from one isomer to another is a reaction event in nearby local minima. dynamics simulation needs to be optimized.
dynamics simulations.
the molecular dynamics simulation.
• The structure of molecules at selected frames within
• A new method for simulation data analysis is developed to Figure 5. Graphs of processed and unprocessed binary time series of bond order.
identify the occurrences of such reactions within simulations. the simulations were optimized to find the lowest
energy structures. • For each pair of atoms, two
• Optimized structures examined to find new isomers. unprocessed time series are made,
Element Legend (for figures to the left)
one from the molecular dynamics
Iron: Pink Carbon: Grey • Optimized structures were also found automatically
Oxygen: Red Nitrogen: Blue trajectory and one from the set of
(Bonds between atoms drawn with though RMSD comparisons.
optimized geometries.
covalent radii thresholds.) • Many isomers were found to be lower in energy than
• The raw time series are processed
those published in the literature.
• The Fe4N(CO)12 iron- using parameters such as bond
Figure 1. Fe4N(CO)12 electrocatalyst (Figure 1) can
Figure 4. Sample theoretical Fe3(CO)9 isomers. order thresholds and the
iron-electrocatalyst. possibly undergo persistency of bond order change.
rearrangement of its carbonyl • The specific values for the
(CO) groups after the loss of a parameters are varied in order to
CO when the electrocatalyst is match the processed time series
twice reduced. made from the molecular
(Published Global • The methodological framework dynamics trajectory to that made
Minimum)
developed in this project will from the optimized geometries.
Figure 2. Sample isomer be used to study theoretical
of Fe3(CO)9.
reaction events for the
electrocatalyst.
• A group of simpler molecules, Objective Functions
the Fe3(CO)9 (Figure 2),
Fe3(CO)10, Fe3(CO)11 (Figure 3), • Multiple processing methods can be used in tandem with each • An objective function is defined to measure the deviation of the
Goals for Detecting Reaction Events in
and Fe3(CO)12 iron carbonyl other in order to process the bond order time series extracted reaction events detected by a processed molecular dynamics
clusters will be used to develop
Molecular Dynamics Simulations from molecular dynamics simulations. bond order time series from the reference set of reaction events.
Figure 3. Sample isomer
of Fe3(CO)11. this analysis tool. • Produce time series of ab initio bond order indices1 • Such methods include specifying bonding thresholds, • The objective function shown in Figure 6 assigns a score to a
extracted from molecular dynamics simulations and recognizing only persistent changes in bond order, and certain combination of bond order threshold and transition gap
geometry optimizations. smoothing the raw bond order time series by filtering out threshold by determining how closely the processed time series
Acknowledgements irregular fluctuations. matches the set of “real” reaction events.
• Apply signal processing methods to the bond order
• The parameters used to process the raw bond order time • The objective function shown in Figure 7 is another way of
time series to detect reaction events that occur
We acknowledge the National Science Foundation for funding the series (the processed time series is shown by the blue lines in assigning a score to a combination of parameters. The function
University of California Davis Chem Energy Research Experience for within simulations.
Figure 6a) are adjusted to maximize the detection of the measures how much of the of the “real” reactions are found as a
Undergraduates program which makes this work possible. We also • Use the methodology developed to find theoretical
reactions in the “gold standard” reaction event time series (red function of the percentage of the trajectory covered.
thank Yudong Qiu for his contributions. isomers of the iron carbonyls, the Fe4N(CO)12 iron- lines in Figure 6b).
electrocatalyst, and other systems of interest. Figure 6a. Figure 6b.
Figure 6. a) An objective
References function where a score is
Figure 7. An objective given by determining the
1. Mayer, I. Bond order and valence indices: A personal function that gives a minimum box size
account. J. Comput. Chem. 2007 (28): 204–221. score by comparing the required to cover all of
DOI:10.1002/jcc.20494. percentage of the the reaction events. b) A
2. Wang, H; Xie, Y; King, R. B.; Schaefer III, H. F. Remarkable molecular dynamics map of the scores made
Aspects of Unsaturation in Trinuclear Metal Carbonyl trajectory covered with for different
Clusters:  The Triiron Species Fe3(CO)n (n = 12, 11, 10, 9). the percentage of the combination of
Journal of the American Chemical Society 2006 128 (35):
“real” transitions found. parameter values.
11376-11384. DOI: 10.1021/ja055223+