GRRM26

Comprehensive reaction pathway exploration — AFIR

GRRM lets users go on a treasure hunt for unknown molecules and reactions: from a single starting structure (an equilibrium geometry), it automatically explores reaction pathways by working together with quantum chemistry calculations and machine learning potentials (MLP). GRRM's development began with the anharmonic downward distortion following (ADDF) method, and its exploration is now driven mainly by the artificial force induced reaction (AFIR) method — GRRM's current core search algorithm. AFIR and ADDF are two independent methods, and together they uncover all four fundamental types of chemical reaction [1], so that a single molecular input can map out the possible products and reaction pathways comprehensively.

(1) A → X (Isomerization)
(2) A → X + Y (Dissociation)
(3) A + B → X (Combining synthesis)
(4) A + B → X + Y (Exchanging synthesis)

New!! Optimization of search performance for high-speed energy calculation programs.

Search performance has been significantly improved when using fast energy and gradient evaluation methods such as semi-empirical methods (SEM) and machine learning potentials (MLP). It is now possible to routinely construct reaction pathway networks containing more than 1,000 pathways using fast methods like SEM and MLP [2][3][4].

Kinetic simulation and kinetics-based navigation — RCMC

Kinetic simulation and kinetics-based navigation using the rate constant matrix contraction (RCMC) method — available since GRRM20 — extend GRRM's automated search to complex reaction pathway networks. The kinetic simulation evaluates the population of chemical species that persist beyond a specified lifetime, while the kinetics-based navigation dramatically speeds up the reaction pathway search (SC-AFIR) by restricting it to kinetically feasible paths under given reaction temperatures and lifetimes [5].

New!! Explicit consideration of concentration effects in RCMC calculations.

Prior to GRRM23, reaction pathway network representations were limited to stoichiometric reactions where all reagents were mixed in equal proportions (the molar ratio of all reagents being 1:1:1…) due to limitations in the computational algorithm. The new algorithm introduced in GRRM26 enables RCMC kinetic simulations that explicitly consider concentration effects, making it possible to apply them to complex systems where catalyst coating rate and reagent concentration affect reaction efficiency [6].

New!! RCMC under a Microcanonical Ensemble condition.

It is now possible to perform RCMC kinetic simulations under a Microcanonical Ensemble. The implemented full-RCMC (f-RCMC) method enables the calculation of final branching ratios for products without performing time evolution of the rate equations. While conventional methods take approximately 95 hours to solve the rate equation for C₄H₅, consisting of 234 elementary processes, f-RCMC has been reported to calculate the branching ratio in less than 1 second [7].

Retrosynthetic analysis — QCaRA

Building on the RCMC method, GRRM performs reverse (retrosynthetic) kinetic analysis across complex reaction pathway networks. It predicts the yields of a target product for all reactions that could lead to it from other species on the network [8]. Used as a kinetics navigation tool, this enables quantum chemistry-aided retrosynthetic analysis (QCaRA): taking only the structure of a product as its input, QCaRA has already correctly identified the reactants of various known reactions, including the synthesis of small natural products [9].

External interface integration

GRRM works together with a range of external computation engines — most notably Gaussian, along with ORCA, GAMESS and others — and, through its general external interface, it is designed to make it easy to connect other solvers that users already have, whether quantum chemistry engines or machine learning potentials (MLP), beyond the officially supported ones. Building on this interface, GRRM can also control structure optimization and exploration with externally developed modules — modifying the search order near local minima, redirecting exploration routes, or applying external bias potentials — and it is the same interface that lets GRRM couple with machine learning potentials. Combined with a rapidly exploring random tree (RRT) algorithm [10] and a graph neural network for path selection [11], it accelerates the reaction pathway search (SC-AFIR) for specific purposes, and it underpins the development and application of virtual ligands for transition-metal catalysts [12].

New!! Improved adaptability to the virtual ligand method and related techniques.

By integrating with the virtual ligand method and its code developed by Specially Appointed Associate Professor Wataru Matsuoka and his colleagues, ligand optimization of organometallic catalysts can be performed. References [13] and source code [14] for the virtual ligand method are provided below.

And many more functions

Beyond the highlights above, GRRM has steadily accumulated a wide range of functions for comprehensive reaction pathway exploration. For a full, side-by-side view of what each version supports, see the comparison table under Functional specs ("See more").

New!! Other performance improvements.

The GRRM source code has been completely redesigned, and numerous performance improvements have been implemented.