MARTINI: Multifaceted Simulation Framework

• MARTINI is a polysemous framework that defines coarse-grained force fields for biomolecular simulations, heavy-ion event generators, neutrino cross-section models, radio-interferometry, and lattice magnetism.
• In molecular simulation, the Martini model maps groups of atoms to coarse-grained beads, enabling efficient exploration of larger time and length scales with maintained chemical accuracy.
• The framework’s adaptability is demonstrated through extensions like Martini 3, specialized carbohydrate and DNA models, and innovations in non-equilibrium and many-body techniques.

to=arxiv_search.search 】【：】【“】【json_string": "{\"query\": \"all:MARTINI force field coarse-grained model event generator neutrino arXiv\", \"max_results\": 10, \"sort_by\": \"submittedDate\", \"sort_order\": \"descending\"}"} สามสิบเอ็ดassistant to=arxiv_search.search_RGCTX 】【。】【”】【json_string": "{\"query\": \"all:Martini model coarse-grained force field materials science arXiv\", \"max_results\": 10, \"sort_by\": \"relevance\", \"sort_order\": \"descending\"}"} to=arxiv_search.search 亂倫json_string": "{\"query\": \"ti:Martini 3 Coarse-Grained Force Field for Carbohydrates OR ti:Martinize2 and Vermouth OR ti:The Martini Model in Materials Science\", \"max_results\": 10, \"sort_by\": \"relevance\", \"sort_order\": \"descending\"}"} MARTINI is a polysemous research term whose principal contemporary usage denotes a family of coarse-grained molecular simulation models, especially the Martini force field and its Martini 3 reparameterization, but it also identifies a heavy-ion event generator, a neutrino–nucleus cross-section model, a radio-astronomical mock-observation package, and a class of lattices in exactly solved spin models (Alessandri et al., 2020, 0909.2037, Oman, 2024, Russo et al., 19 Aug 2025, Zad et al., 2022). In the molecular-simulation literature, Martini is a building-block coarse-grained framework initially developed for biomolecular simulations and subsequently extended to materials science, polymers, carbohydrates, membranes, and hybrid soft-matter systems (Alessandri et al., 2020).

1. Name, scope, and principal usages

The term “MARTINI” does not denote a single formalism across all disciplines. In current arXiv literature, it labels several unrelated but internally coherent frameworks.

Usage	Field	Defining description
Martini / Martini 3	Coarse-grained molecular simulation	Building-block coarse-grained force field for biomolecules and soft materials (Alessandri et al., 2020)
MDPD-MARTINI	Soft-matter simulation	MARTINI “Lego” approach transferred to many-body dissipative particle dynamics (Carnevale et al., 2023)
MARTINI	Heavy-ion collisions	Modular Algorithm for Relativistic Treatment of heavy IoN Interactions (0909.2037)
Martini-Ericson-Chanfray-Marteau	Neutrino interactions	RPA-based (anti)neutrino cross-section model implemented in GENIE (Russo et al., 19 Aug 2025)
MARTINI	Radio astronomy	Modular Python package for synthetic 21-cm HI interferometry (Oman, 2024)
martini and martini-diced lattice	Statistical mechanics	Lattice geometries in exactly solved Ising-Heisenberg models (Zad et al., 2022)

Within molecular simulation, Martini has become a general-purpose coarse-grained platform rather than a lipid-only or biomolecule-only model. The materials-science review explicitly emphasizes that the model’s building-block principle does not pose restrictions on its application beyond biomolecular systems, and highlights polymers, nanoparticles, organic electronics, ion-conducting materials, supramolecular assemblies, ionic liquids, and oils or bitumen as established application classes (Alessandri et al., 2020).

2. Martini as a coarse-grained force-field family

The Martini model is a coarse-grained force field in which, typically, four heavy atoms and their bonded hydrogens are mapped to a single coarse-grained bead, thereby reducing particle count and extending accessible time and length scales relative to all-atom MD (Alessandri et al., 2020). Its canonical bead taxonomy captures polarity, charge state, and hydrogen-bonding character; the review cites Polar (P), Nonpolar (N), Apolar (C), and Charged (Q) classes, 18 basic bead types, and additional “small” and “tiny” beads for aromatic rings and other highly resolved groups (Alessandri et al., 2020). Non-bonded interactions are usually represented with a 12-6 Lennard-Jones form,

$$V_{LJ}(r)=4\epsilon\left[\left(\frac{\sigma}{r}\right)^{12}-\left(\frac{\sigma}{r}\right)^6\right],$$

with bonded terms supplied by standard classical bonds, angles, and dihedrals (Alessandri et al., 2020).

A defining principle is transferability through a building-block, or “Lego,” philosophy. In the MDPD transfer paper, intermolecular interactions between coarse-grained beads representing chemical units of different polarity are obtained through water–octanol partition coefficients, enabling a finite bead library to be recombined into larger molecules (Carnevale et al., 2023). The same logic underlies later automation work in Martini 3, whose broader bead set—particularly for small molecules—expands chemical resolution while making mapping more context dependent (Bigting et al., 14 Nov 2025).

Martini 3 is described as a full re-parametrization of the Martini coarse-grained model for biomolecular simulations, with improved interaction balance and a broader set of bead types and “labels” intended to improve miscibility, self-assembly, packing, and transferability across materials-science use cases (Grünewald et al., 2022, Alessandri et al., 2020). The materials-science review also associates Martini 3 with improved molecular packing, improved interaction balance based on free energies and miscibility data, and better support for high-throughput workflows through tools such as AutoMARTINI, Cartographer, Swarm-CG, Martinize 2, and Polyply (Alessandri et al., 2020).

3. Mapping, topology generation, and faster dynamical formalisms

The contemporary Martini ecosystem couples force-field design to explicit mapping algorithms and topology-generation software. For proteins and heterogeneous biomolecular systems, Martinize2 and the Vermouth library provide a six-stage pipeline—parsing input, identify and repair, resolution transformation, create topology, post-processing, and write output—and add automatic handling of protonation states, post-translational modifications, elastic-network tuning, and non-protein molecules such as ligands (Kroon et al., 2022). Their high-throughput benchmarks are unusually large: 73% of 87,084 I-TASSER structures were processed successfully end-to-end, and 199,993 of 200,000 randomly selected AlphaFold structures proceeded successfully through conversion and minimization (Kroon et al., 2022).

For small molecules, “Martini Mapper” formalizes a fragment-based Martini 3 workflow that starts from canonical SMILES, partitions molecules into aromatic rings, non-aromatic rings, and acyclic fragments, and exports GROMACS-compatible .gro and .itp files (Bigting et al., 14 Nov 2025). The framework generated Martini 3 models for more than 5,000 molecules across four chemically diverse datasets, mapped molecules containing up to 126 heavy atoms, and benchmarked a curated subset of 1,081 structures by octanol-water free-energy and $\log P$ calculations; the reported results include a final $R^2=0.83$, RMSE $=0.99$ for the Bereau dataset after bead refinement, $R^2=0.63$, RMSE $=0.84$ for a Kaggle dataset, and $R^2=0.71$, RMSE $=0.79$ for a 2D benchmark (Bigting et al., 14 Nov 2025).

Carbohydrate modeling illustrates the same trend toward explicit, transferable mapping rules. The Martini 3 carbohydrate work introduces a canonical mapping scheme that decomposes mono-, oligo-, and polysaccharides into recurrent fragments, scales bond lengths uniformly by 15% over naïve center-of-geometry distances,

$$l_{\text{Martini 3}} = 1.15 \times l_{\text{COG-map}},$$

and reports accurate osmotic pressures for mono- and disaccharide solutions at low to moderate concentrations, correct differentiation between dextran and cellulose solubility, and transferability to glycolipids and membrane–protein binding problems (Grünewald et al., 2022).

A separate line of development replaces conventional MD with many-body dissipative particle dynamics while preserving Martini’s bead-level chemistry. In MDPD, the conservative force is written as

$$F_{ij}^C = A_{ij}\omega^C(r_{ij})\hat{e}_{ij} + B\left(\bar{\rho}_i + \bar{\rho}_j\right)\omega^D(r_{ij})\hat{e}_{ij},$$

with a density-dependent repulsive term added to a bead-specific attractive term (Carnevale et al., 2023). The initial MDPD-MARTINI study reported bilayer formation 4–7 times faster than standard MARTINI MD for DPPC while retaining comparable area per lipid, bilayer thickness, undulation spectra, and lipid order parameters (Carnevale et al., 2023). A later force-field paper extended this to DPPC, POPC, and DOPC bilayers of 512 and 8192 lipids, reporting deviations of 1–2 Å$\log P$0 in area per lipid, bilayer-thickness differences below 2 Å, correct $\log P$1 long-wavelength undulation behavior, and “an order of magnitude” speed-up, with minutes of MDPD simulation reaching phases that require hours in MD MARTINI (Kramarz et al., 2 Jan 2025).

4. Molecular applications and empirical performance

The Martini force-field family now spans a broad molecular-design and materials-simulation space. The materials-science review catalogs hydrogels, polymer brushes, dendrimers, block copolymers, polymer coatings and interfaces, nanocomposites, carbon nanomaterials, gold nanoparticles, organic-electronics blends such as P3HT:PCBM, ion-conducting materials including PEDOT:PSS and Nafion, supramolecular polymers, ionic liquids, and asphaltene aggregation as established application areas (Alessandri et al., 2020). This breadth is reinforced by specialized transferable models developed within Martini itself.

For poly(ethylene oxide), a new transferable Martini model was parameterized against eight free energies of transfer of dimethoxyethane, the radius of gyration in water at high dilution, and angle and dihedral distributions from atomistic simulations (Grunewald et al., 2019). The resulting model was validated in five areas: densities and phase behavior of small oligomers and water mixtures, chain dimensions in water, diglyme, and benzene over 1.2–21 kg/mol, PEGylated-lipid bilayers in brush and mushroom regimes, phase behavior of several PEO-based nonionic surfactants, and compatibility with an existing MARTINI PS model for PS/PEO block copolymers (Grunewald et al., 2019). For poly($\log P$2-peptoids), a MARTINI-compatible backbone-and-sidechain construction based on polysarcosine was reported to predict hydration free energies of other peptoids, radii of gyration over a wide range of chain lengths, and a sequence-defined diblock peptoid coil–globule transition in binary solvent mixtures (Gao et al., 2019).

Martini 3 has also been adapted to functional molecular devices. In the bistable-nanomachine study, coarse-grained PNIPA and pyridine–furan oligomers were used to reproduce thermally activated spontaneous vibrations, stochastic resonance, and force- or temperature-driven bistability analogous to Euler arches and Duffing oscillators, with end-to-end distance as the principal collective variable and oscillating external fields used to probe resonance behavior (Muratov et al., 18 Jul 2025). A plausible implication is that Martini’s utility now extends beyond equilibrium self-assembly to nonequilibrium rare-event and response problems, provided the mapping and bonded terms are tuned to the targeted phenomenology.

5. Limitations, corrections, and domain-specific reparameterizations

Martini’s central methodological promise is transferability, but the same literature repeatedly shows that transferability is conditional rather than automatic. The polysaccharide re-evaluation found that default MARTINI drastically overestimates aggregation propensity in aqueous solution: glucose, sucrose, cyclodextrins, and more complex glycans aggregated in simulation below their solubility limits, and the authors traced this to an imbalance among solute–solute, solute–water, and water–water Lennard-Jones interactions (Schmalhorst et al., 2017). Their corrective proposal was a uniform scaling of saccharide–saccharide Lennard-Jones well depths,

$\log P$3

with an optimal $\log P$4 around 0.5 for the antifreeze water model; this abolished nonphysical aggregation at physiologically relevant concentrations without disrupting other reported properties (Schmalhorst et al., 2017). Martini 3 carbohydrate parametrization can be read as a later, more systematic response to the same class of issues, because it explicitly targets osmotic pressure, solubility, and conformational fidelity rather than relying on legacy carbohydrate mappings (Grünewald et al., 2022).

For DNA and DNA nanostructures, the comparison with oxDNA and all-atom MD shows a similar pattern. Martini with a soft elastic network gives persistence lengths and stretch moduli for dsDNA, PX DNA, and DNA nanotubes in very good agreement with atomistic and experimental reference ranges, whereas a stiff elastic network yields order-of-magnitude overestimates of rigidity (Naskar et al., 2021). The same study concludes that Martini models proved inadequate to capture salt concentration effects on mechanical properties with increasing salt molarity, and that time-evolved PX DNA and DNA nanotube structures from oxDNA, unlike Martini, are comparable to all-atom MD structures (Naskar et al., 2021).

Solid-state and interfacial systems impose additional constraints. In ZIF-8, Martini 2.0 and 3.0 coarse-grained models reproduce overall structure and generally reproduce lattice parameters, but Martini 2.0 tends to overestimate the cell parameter and none of the tested models captures the guest-induced swing effect within the scope of MD simulations (Alvares et al., 2023). In thermoplastic starch nanocomposites, the liquid–liquid partitioning-based MARTINI-2 force field shows freezing and compaction of polymer chains near a TMA-MMT clay surface; the reported solution is a rescaling of the dispersive component of TPS–MMT cross-interactions constrained by all-atom structural, thermodynamic, and dynamic observables (Patidar et al., 2024). In water at high temperature, the classical Martini water model was deemed unsuitable for phase change and heat transfer, motivating MARTINI-E, a reparameterized water model optimized with a Genetic Algorithm, an Artificial Neural Network, and Nelder–Mead; MARTINI-E accurately reproduces density, enthalpy of vaporization, and surface tension at $\log P$5C and was validated by energy conservation and latent-heat tests in a lamellar system (Yesudasan, 2019).

Taken together, these results make a recurrent point. Martini is not a fixed universal closure; it is a coarse-grained design language whose reliability depends on property-driven calibration, careful mapping, and, in some domains, explicit correction of cross-interactions or water models.

6. High-energy, neutrino, and radio-astronomical MARTINI frameworks

Outside molecular simulation, MARTINI appears in high-energy phenomenology as the Modular Algorithm for Relativistic Treatment of heavy IoN Interactions, a comprehensive event generator for hard and penetrating probes in relativistic heavy-ion collisions (0909.2037). Its core components are a time evolution model for the soft background, PYTHIA 8.1 for initial hard scattering and hadronization, and the McGill-AMY parton-evolution scheme for radiative and elastic in-medium energy loss (0909.2037). In a recent jet-quenching study, the same framework was used with event-by-event hydrodynamical backgrounds from MUSIC with IP-Glasma initial conditions to show that shower formation time after the initial hard scattering is essential for a simultaneous description of charged hadron and jet $\log P$6, and that jet shape and fragmentation-function ratios are sensitive to whether leading-order, next-to-leading-order, or non-perturbative collision kernels are used (Modarresi-Yazdi et al., 2024).

In neutrino–nucleus scattering, “MARTINI” denotes the Martini-Ericson-Chanfray-Marteau RPA-based model. Its GENIE implementation covers quasielastic 1p1h and multinucleon 2p2h and 3p3h excitations within a local Fermi gas description with RPA response functions (Russo et al., 19 Aug 2025). The implementation uses lookup tables for hadron tensors on $\log P$7C, $\log P$8O, and $\log P$9Ca, extends to argon through scaling prescriptions, and is validated by direct comparison with original calculations and by comparison to T2K and MicroBooNE measurements, where the paper reports reasonable agreement (Russo et al., 19 Aug 2025). The same source notes limitations: tensors are calculated only for isoscalar targets, removal-energy effects are not explicitly included in the lepton kinematics, and the current “3p3h” implementation yields two outgoing nucleons (Russo et al., 19 Aug 2025).

Astronomical usage is again distinct. MARTINI, in “Mock Array Radio Telescope Interferometry of the Neutral ISM,” is a modular, object-oriented Python package that converts smoothed-particle hydrodynamics simulations of galaxies into synthetic spatially and spectrally resolved observations of the 21-cm HI line (Oman, 2024). Its workflow is decomposed into submodules for the data cube, source galaxy, beam pattern, noise, spectral model, and SPH kernel, each exposed as a class that can be configured or subclassed; a main Martini class then assembles a mock observation by orchestrating these components (Oman, 2024). Here the term designates not a force field or transport model but a simulation-to-observation interface.

7. Martini lattices and cross-disciplinary usage

In exactly solvable statistical mechanics, the martini and martini-diced lattices are geometrical settings for a spin-1/2 Ising-Heisenberg model solved by a star-triangle transformation that maps the system onto an effective spin-1/2 Ising model on a triangular lattice (Zad et al., 2022). The model exhibits three ground states: a classical ferromagnetic phase with fully saturated Ising and Heisenberg magnetizations, a quantum ferromagnetic phase in which the Heisenberg magnetization is reduced to one-third of saturation, and a macroscopically degenerate frustrated disordered phase (Zad et al., 2022). The reported residual entropies are $R^2=0.83$0 for the martini lattice and $R^2=0.83$1 for the martini-diced lattice, and all three phases coexist at a triple point at $R^2=0.83$2 (Zad et al., 2022). The paper’s central physical point is that frustration can emerge not only from antiferromagnetic interactions but also from competing ferromagnetic Ising and Heisenberg terms of easy-axis and easy-plane type, respectively (Zad et al., 2022).

This cross-disciplinary record suggests a terminological convergence around modularity, reduction, or composite structure, rather than a shared mathematical lineage. In molecular simulation, Martini is a transferable coarse-graining strategy; in heavy-ion and neutrino physics, it is a process generator or response model; in astronomy, a modular synthetic-observation package; and in lattice magnetism, a geometric label. The common name therefore identifies several mature research programs, not a single unified method.