Microscopic Mechanism of the Thermal Amorphization of ZIF-4 and Melting of ZIF-zni Revealed via Molecular Dynamics and Machine Learning Techniques

Technical Summary: Microscopic Mechanism of the Thermal Amorphization of ZIF-4 and Melting of ZIF-zni

Problem Statement
Amorphous Metal-Organic Frameworks (MOFs) offer a unique combination of intrinsic porosity and mechanical robustness, yet the microscopic mechanisms governing their formation from crystalline precursors remain poorly understood. Specifically, the ordered–disordered phase transitions in Zeolitic Imidazolate Frameworks (ZIFs), such as the thermal amorphization of ZIF-4 and the melting of ZIF-zni, present significant challenges for computational modeling. While experimental data exists, questions regarding the specific mechanisms of amorphization, the nature of intermediate states, and the behavior beyond first-neighbor distances remain unanswered. Standard ab initio molecular dynamics (AIMD) are often limited by system size and timescales, while traditional force fields frequently fail to capture the necessary metal–ligand reactivity required to model bond breaking and formation during these transitions.

Methodology
The authors employed a hybrid approach combining classical molecular dynamics (MD) with machine learning (ML) to investigate the phase transitions of two Zn(C3H3N2)2 polymorphs: porous ZIF-4 and dense ZIF-zni.

1. Force Field: Simulations were conducted using the nb-ZIF-FF (non-bonded ZIF-FF) force field. This reactive force field utilizes Morse potentials to model metal–ligand bonds, allowing for bond breaking and formation. It incorporates a "dummy atom" model to enforce the correct tetrahedral topology around Zn²⁺ centers, ensuring accurate reproduction of structural, mechanical, and thermodynamic properties for crystalline, amorphous (ZIF_a), and liquid (ZIF_liq) phases.
2. Structure Generation: Amorphous structures (ZIF_a) were generated via simulated annealing (heating to 900 K and cooling) from both ZIF-4 and ZIF-zni precursors. The resulting structures were validated against experimental density, bulk modulus, and radial distribution functions (RDFs).
3. Machine Learning Classification: To distinguish between the four phases (ZIF-4, ZIF-zni, ZIF_a, ZIF_liq) at the local atomic level, the authors trained a neural network.
   • Descriptors: Behler-Parrinello Symmetry Functions (BPSF) were computed for Zn²⁺-centered environments, serving as unbiased inputs.
   • Architecture: A neural network with 12 input nodes (symmetry functions), one hidden layer (6 nodes), and 4 output nodes (phase probabilities).
   • Performance: The model achieved 90.3% accuracy in classifying local environments, successfully distinguishing between the two disordered phases (ZIF_a and ZIF_liq) based solely on structural information.
4. Simulation Protocol: Large-scale NPT MD simulations were performed using LAMMPS. Transition temperatures were determined by identifying equilibrium conditions where phases coexist without net growth of one over the other.

Key Results

• Thermodynamics: The study determined equilibrium transition temperatures of 420 K for ZIF-4 amorphization and 625 K for ZIF-zni melting. While these values differ from experimental observations (589 K and 863 K, respectively), the model correctly reproduces the thermodynamic stability trends, confirming ZIF-4 is metastable at ambient conditions. Calculated enthalpies and entropies align well with reference values.
• Mechanism of ZIF-4 Amorphization: The transition from crystalline ZIF-4 to amorphous ZIF_a is a two-step process:
  1. Formation of a low-density, liquid-like intermediate phase (ZIF_liq-like).
  2. Transformation of this intermediate into the high-density amorphous phase (ZIF_a).
     The process involves a decrease in connectivity followed by a reorganization. The formation of the amorphous phase is anisotropic, propagating faster along the x-axis. This anisotropy correlates with local density distributions and the lability of coordination bonds; bonds oriented in the x-direction are more prone to breaking, facilitating connectivity exchange.
• Mechanism of ZIF-zni Melting: The melting of ZIF-zni also follows a two-step pathway, but in reverse order of density evolution compared to ZIF-4:
  1. Formation of an amorphous-like phase from the crystal.
  2. Transition to the liquid phase (ZIF_liq).
     The liquid phase nucleates within the bulk of the intermediate disordered phase.
• Microscopic Triggers: The initiation of amorphization is linked to specific connectivity changes rather than simple bond breaking. While tricoordinated Zn defects occur frequently, they are reversible. The critical event triggering the phase transition is the formation of a new bond between a Zn²⁺ center and a different ligand molecule (changing the network connectivity), rather than a simple rotation or reformation of the original bond. This event is accompanied by a sudden increase in the population of liquid-like environments identified by the neural network.
• Structural Evolution: The formation of 3-membered rings (Zn–ligand–Zn) is observed to correlate with the growth of the disordered phase, suggesting these topological features are characteristic of the amorphous state, whereas the crystal primarily contains 4, 6, and 8-membered rings.

Significance and Claims
The paper claims to provide a detailed microscopic understanding of thermal amorphization and melting in ZIFs, addressing gaps left by previous studies that ignored intermediate states or lacked sufficient system size. The primary contributions are:

1. Validation of nb-ZIF-FF: Demonstrating that this force field can accurately reproduce the structures and properties of crystalline, amorphous, and liquid ZIF phases, including subtle angle distributions that other reactive force fields (like ReaxFF) fail to capture.
2. ML-Augmented Dynamics: Establishing a methodology where a neural network, trained on symmetry functions, can track the time-evolution of phase transitions with high spatial resolution, distinguishing between structurally similar disordered phases.
3. Mechanistic Insight: Identifying that both amorphization and melting are two-step processes involving intermediate phases with distinct densities and that the amorphization of ZIF-4 is an anisotropic process driven by the lability of specific coordination bonds.

The authors conclude that their work offers a framework for studying reactive processes in MOFs, potentially aiding the design of amorphous MOFs for industrial applications such as water purification and catalysis, though they do not propose new experimental protocols or specific future applications beyond the scope of understanding the fundamental mechanisms.