How to Choose Molecular Dynamics (MD) Simulation Software: 6 Key Points and 14 Tools Compared

In the field of materials development, we often encounter atomic-level questions such as, "Why do ions diffuse easily in this material?", "What is happening near the interface?", and "How do defects affect the properties?". A powerful tool for answering these questions is "molecular dynamics (MD) simulation," which visualizes the movement of atoms and molecules through calculations.

Performing MD simulations requires software, but there are many types of MD simulation software available, each with different strengths in terms of the type of system they are best suited for, calculation speed, ease of use, and licensing models. Therefore, many people may be unsure of what to choose.

This article provides a clear and concise explanation in a table format for researchers in the materials field who are considering introducing MD simulation software. It covers essential knowledge to consider before selecting software, six key points to look at during the selection process, and an overview of 14 representative MD simulation software programs. Reading this article should give you hints on how to identify software that suits your company's research and development themes.

What is MD Simulation? | Essential Knowledge for Software Selection

Before explaining how to choose MD simulation software, we will first explain the basic knowledge of MD simulation necessary for software selection. If you do not need this explanation, please proceed to "Points to consider when choosing MD simulation software."

Basic concepts of MD simulation

MD simulation is a computational method that solves Newton's equations of motion for each individual atom that makes up a substance, and tracks how the atoms move along the time axis. Specifically, it calculates the "forces" acting on the atoms and repeatedly updates the position and velocity of the atoms little by little according to those forces. This allows us to obtain the position, velocity, and energy of the atoms at each point in time [1].

What can and cannot be done with MD simulation

The strength of MD simulation lies in its ability to capture "movement" and "how changes occur" at the atomic and molecular level. For example, MD simulation is extremely useful for visualizing and analyzing phenomena that are difficult to observe through experiments alone, such as the following:

• Diffusion: How ions and molecules move within a material.
• Interface behavior: The arrangement and behavior of atoms near the interface where different materials come into contact.
• Behavior near defects: How defects in crystals (vacancies, interstitial atoms, etc.) affect material properties.
• Adsorption and Desorption: How molecules adsorb to and desorb from a surface.

On the other hand, MD simulations require calculations to be performed in very short time increments of about 1 to a few femtoseconds (10⁻¹⁵ seconds) to avoid missing the rapid movements of atoms. If the time increment is too large, it becomes impossible to accurately track the movements of atoms, and the overall accuracy of the simulation drops significantly. Therefore, there are practical limitations to the total time scale that can be handled. For example, phenomena that proceed slowly on a scale of seconds to years, such as the long-term degradation of materials, or phenomena that occur only rarely, such as crystal nucleation, are considered difficult to observe directly with conventional MD simulations [2].

Three methods for calculating forces in MD simulation software

As explained earlier in "Basic Concepts of MD Simulation," MD simulations require the calculation of "forces" acting on atoms at each step. In fact, this force calculation is a crucial process that accounts for the majority of the overall computation time of an MD simulation, and the accuracy, speed, and scale of the system that can be handled vary greatly depending on the method used to determine the forces [1].

One common method for determining force involves pre-designing a function based on experience that represents the relationship between atomic arrangement and energy. This function indicates "how much potential energy an atom possesses in a given arrangement," and the force acting on each atom can be calculated from the function's slope (the derivative with respect to atomic coordinates). Such a function is called a "potential" or "force field."

On the other hand, instead of humans designing these functions in advance, there is also a method that calculates the electronic state of the atom at each step and then determines the forces acting on the atom from the results.

Due to these differences in force calculation methods, MD simulations can be broadly classified into the following three methods.

Classical MD (Method of calculating force: Classical force field)

This method calculates forces using force fields optimized for specific materials or molecular systems. Because it does not consider electronic states at each step of the simulation and calculates forces using only empirically designed force fields, the force fields used in this method are specifically called "classical" force fields. The calculation speed is very fast, and under certain conditions, it is possible to perform long-running simulations on the order of microseconds (^10⁻⁶ seconds) for large systems of hundreds of thousands to millions of atoms [1]. Although it is called "classical," it remains the most widely used method, mainly in materials science and biological systems.

First-principles MD (Force calculation method: First-principles calculation)

Of the three methods, "first-principles MD" is the only one that does not use a predetermined potential function, but instead directly derives forces from the electronic state at each step. First-principles calculations (such as DFT calculations) are performed at each step to determine the forces acting on atoms based on their electronic states. The advantage is that, because electronic states can be taken into consideration, chemical reactions and the formation and breaking of bonds are easier to handle in principle [3]. On the other hand, the computational cost of first-principles calculations is very high, so the systems that can generally be handled are limited to around a few hundred atoms, and the time scale is limited to the order of picoseconds (^10⁻¹² seconds) [1].

(For more information about DFT, please see our blog post "[Beginner's Guide] What is Density Functional Theory (DFT)? | Basics")

Machine Learning Potential (MLIP) for MD (Force Calculation Method: MLIP)

This method predicts the forces acting on atoms using a machine learning model that has learned the relationship between atomic arrangement and energy/force. Such a machine learning model is called a "Machine Learning Interatomic Potential (MLIP)" or "Machine Learning Force Field." In this article, MD simulations using MLIP will be referred to as "MLIP-MD."

Machine learning models are primarily trained using data obtained from first-principles calculations such as DFT. Therefore, it is possible to determine forces with near-perfect accuracy and at a faster speed, without having to perform first-principles calculations every time, as explained earlier with first-principles MD. Recently, simulations involving tens of thousands of atoms and tens of nanoseconds (nanosecond: ^10⁻⁹ seconds), which could not be handled by conventional first-principles calculations, have become practical [2].

On the other hand, the accuracy of MLIP depends on the scope and quality of the training data, so the prediction accuracy may decrease for structures and conditions not included in the training data. Therefore, it is important to select an MLIP that corresponds to the target material system or phenomenon [4].

(For more information on MD simulation and MLIP, please also see our blog articles "Introduction to Molecular Dynamics Simulation" and "Introduction to Machine Learning Potential: Simulation Technology to Accelerate Materials Development.")

Key points to consider when choosing MD simulation software

When choosing MD simulation software, it's crucial not to rely solely on the software's name or reputation, but to consider your company's research themes and implementation environment. Here are six key points to consider when selecting MD simulation software.

Point 1: Does it correspond to the materials systems and phenomena you are interested in?

The first thing you should check is whether the software supports the force fields and potentials relevant to your research topic. Generally, certain force fields and potentials are designed for specific material systems (metals, semiconductors, polymers, etc.) and cannot be directly applied to different material systems. The types of force fields included vary depending on the MD simulation software, so prior confirmation is crucial.

Another important point is whether the software has the necessary methods implemented to calculate the physical properties or phenomena you want to observe. For example, if the software has analysis functions that correspond to what you want to investigate, such as calculating the diffusion coefficient, analyzing the interface structure, or evaluating material strength, you can avoid the trouble of creating analysis scripts from scratch yourself.

Point 2: Can it handle the size and time scale of the system you want to work with?

It's also important to consider whether the results can be obtained within a realistic computation time for the size of the system you want to work with (number of atoms) and the time scale you want to observe. This involves both the "force calculation method" and the "software's ability to handle speed optimizations."

As mentioned in "Three Methods for Calculating Forces in MD Simulation Software," classical MD, first-principles MD, and MLIP-MD each have different system sizes and time scales they can handle. Therefore, it is best to first estimate "what system size and time scale are needed for the phenomenon you want to observe," choose a force calculation method that can handle those conditions, and then select software that supports that method.

Furthermore, even with the same force calculation method, the calculation speed varies depending on the software. For example, for calculations involving large-scale systems, software that supports GPUs or can perform parallel computing can execute calculations much faster.

Point 3: Is it highly scalable?

Scalability refers to the degree of freedom users have to add new features to the software or customize existing ones. For example, in the case of classical MD software, it's important whether new force fields and potentials can be added, and in the case of first-principles MD software, it's important whether new electronic state calculation methods and analysis functions can be added.

Generally, open-source MD simulation software offers high extensibility. Because their source code is publicly available, it's easy to add and modify features, and many such programs have active community development of plugins and extension packages. On the other hand, some commercial software also allows integration with external visualization and analysis tools, and customization of calculation procedures.

Point 4: Status of MLIP support

In recent years, the use of MLIP has been rapidly expanding, and existing classical MD software and first-principles MD software are increasingly supporting MLIP. The meaning of "support" differs depending on the software; for first-principles MD software, it means "the ability to export DFT data calculated by the software in a format usable for MLIP training," while for classical MD software, it means "the ability to incorporate externally trained MLIPs into the software and perform MD."

Currently, a workflow is becoming common in which an external MLIP is trained using the results of DFT calculations performed with first-principles MD software, and then the trained MLIP is integrated into classical MD software to perform MLIP-MD. Furthermore, software with built-in MLIPs that can perform DFT calculations, MLIP training, and MLIP-MD execution in a single, integrated process is also emerging.

Point 5: Is it easy to implement and easy to continue using within the company?

It's necessary to examine not only the technical aspects of the software, but also its implementation and operation.

First, when considering implementation, you should comprehensively evaluate the software's licensing model (open source or commercial), purchase costs, and the necessary computing environment. For example, even with free software, large-scale calculations may require a high-performance CPU or GPU and large-capacity storage, potentially resulting in higher total implementation costs than anticipated. If you cannot prepare your own computing environment, choosing cloud-based software with a low environment setup burden is a viable option.

To ensure continued use within a company after implementation, the availability of documentation and learning resources is crucial. Many open-source MD simulation software programs have extensive forums and tutorials run by user communities, while commercial software offers vendor support and training. Generally, software with a large user base and extensive usage history has more information available, making it easier to find solutions when problems arise.

Point 6: How much expertise in IT and computational chemistry is required?

If you are not familiar with IT or computational chemistry, be sure to check what level of expertise is required to use the software.

Software with a robust GUI allows users to visually perform tasks from initial structure construction to MD simulation execution and result analysis, making it relatively easy to start using even without advanced IT skills. On the other hand, software without a GUI relies on command line or Python scripting, traditionally requiring some knowledge of Linux and programming. However, in recent years, the emergence of large-scale language models (LLMs) such as ChatGPT and Claude, and coding agents, has significantly lowered the barrier to scripting. In fact, from the perspective of compatibility with AI agents, software that can be operated via scripts is easier to integrate with than software that uses a GUI. When choosing software in the future, it would be wise to consider compatibility with LLMs and coding agents as an important factor.

The required knowledge of computational chemistry also varies depending on the software. For example, some software allows you to start MD simulations immediately by simply selecting a built-in force field, while others require the user to finely adjust the force field parameters themselves. Furthermore, the extent to which the software assists in creating the initial structure used in the simulation, such as creating surface and interface models, introducing defects, and arranging molecules, is also an important point.

[Comparison Table] 14 Representative MD Simulation Software Selections

Below, we compare 14 representative MD simulation software programs in a table format, based on the selection points introduced in the previous section. This time, in addition to software for materials science, we have also included software for biomolecules, which you often see mentioned, for reference. Please check which MD simulation software would be suitable for your company.

(※For details and the latest information on each software, please check the developer's official website.)

Regarding "scalability"

Among classic MD software, LAMMPS and OpenMM are particularly well-suited for developing new force fields and potentials. LAMMPS allows for easy addition of features through programming (C++), while OpenMM has a mechanism that automatically calculates forces once the potential is defined. Both software programs are widely used by researchers who want to quickly test new force fields and potentials.

In first-principles MD software, Quantum ESPRESSO is designed to easily add various functions as modules, leading to active community development and feature additions. VASP has a wealth of external tools and plugins that can be combined with the software itself, but since it is not open source, there are restrictions on modifying the software itself.

- Regarding "MLIP compatibility status"

For execution environments of MD (Master Simulation), a widely adopted method uses a Python-based framework called ASE (Atomic Simulation Environment) as a common interface, allowing for easy access to various MLIPs (Multi-Level Programming Initiatives) and execution of MD. On the other hand, LAMMPS and OpenMM also have mechanisms for directly integrating MLIPs into their software, making it possible to execute MLIP-MD without going through ASE. Furthermore, GROMACS and other similar systems are also beginning to support the integration of such MLIPs.

VASP, a software capable of electronic structure calculations and first-principles modeling (MD), has a built-in MLIP (Multiple Modeling Programming), allowing MD to be performed while learning and updating MLIP in real time during simulation. Quantum ESPRESSO, SIESTA, and OpenMX, among others, can learn MLIP based on their own DFT data via external tools.

Matlantis is an atomic-level simulation platform that allows for easy execution of MLIP-MD across a wide range of target systems.

As we have seen, MLIP has recently attracted attention as a technology that aims to achieve both accuracy and speed in MD simulations. However, conventional MLIP-MD also has the following challenges.

• In many cases, a dedicated MLIP (Multiple Life Process Indicator) must be prepared for each target system.
• We need to prepare the large amount of DFT data required for MLIP training ourselves.
• Building workflows to integrate MLIP with first-principles calculation software and classical MD software requires expertise and resources in IT and computational chemistry

One new approach that is attracting attention to solve these problems is "Matlantis," which we offer.

Matlantis is a cloud-based simulation environment that incorporates "Matlantis PFP," a general-purpose MLIP (Mechanical Research Injection Program) capable of handling a wide range of material systems, including catalysts, batteries, semiconductors, polymers, and alloys. Its key feature is that users do not need to prepare a dedicated MLIP for each target system, making it easy to start MD (Mechanical Distortion) simulations with the desired material system. It eliminates the need for processes such as preparing DFT data, training the MLIP, and integrating with other software, significantly simplifying the workflow to start MLIP-MD. The MD simulation itself can be executed by calling PFP from widely used MD frameworks such as ASE and LAMMPS, making it easy for users familiar with these existing tools to migrate. Furthermore, since calculations are completed in a cloud environment, it can be used from a browser without dedicated computing equipment. In May 2026, a function to integrate Matlantis with coding agents such as Claude Code and Codex was added, enabling the automatic generation of simulation code on Matlantis from natural language instructions (press release).

If you'd like to learn more about Matlantis, please see the following.

• For those who want to learn about specific ways to use Matlantis and the benefits of implementing it:Customer Case Studies Page
• For those who want to learn more about Matlantis, including its features and overview:Product documentation download page

summary

This article explains how to choose MD simulation software for materials development, covering essential knowledge to consider before selecting software, six key points to look at during selection, and the characteristics of 14 representative software programs.

When choosing MD simulation software, it's important to select software that suits your research topic (materials/phenomena), required scale, and installation/operation environment, rather than simply choosing a "famous software." Please use the software selection points and comparison table introduced in this article as a reference when choosing software.

[References]

[1] Matlantis blog, "Introduction to Molecular Dynamics Simulation": https://matlantis.com/ja/resources/blog/md-intro/

[2] GC Sosso et al., “Crystal Nucleation in Liquids: Open Questions and Future Challenges in Molecular Dynamics Simulations” Chem. Rev. (2016).

[3] ME Tuckerman, “Ab initio molecular dynamics: Concepts, recent developments, and future trends” PNAS (2005).

[4] R. Jacobs et al., “A practical guide to machine learning interatomic potentials – Status and future” Curr. Opin. Solid State Mater. Sci. (2025).