Uncovering the Role of Critical Bonds in the Thermomechanical Response of Zn₂(1,4-benzenedicarboxylate)₂(1,4-diazabicyclo[2,2,2]octane) Using <i>Ab Initio</i> Simulations and Physics-Constrained Neural Networks

Metal-organic frameworks (MOFs) are susceptible to harsh conditions involving high temperatures, mechanical loading, host/guest chemical interactions, or a combination thereof due to the disruption of underlying bonds. Here, we probed the thermomechanical response of Zn₂(BDC)₂(DABCO) (BDC = 1,4-benzenedicarboxylate and DABCO = 1,4-diazabicyclo[2,2,2]octane) MOF or Zn-DMOF, whose metal nodes are connected to two BDC and DABCO linkers via Zn-O and N-Zn bonds, respectively. We have examined the relative contributions of such bonds toward Zn-MOF's thermomechanical response at 200 and 300 K, by training a physics-constrained neural network with <i>ab initio</i> molecular dynamics (AIMD)-derived interatomic vibration data, and quantitatively estimated interatomic bond strengths within a local environment in the MOF body. We quantitatively show that N-Zn and Zn-O bonds near the Zn-based metal nodes are weaker than their surroundings. These findings were followed up by separately annealing Zn-DMOF at 920 K-AIMD (above its thermal decomposition temperature), mechanically straining it to failure using 0 K-DFT, and examining host/guest interactions in aqueous and acidic environments with 300 K-AIMD. Together, they indicated that structural instability in Zn-DMOF was initiated by the disruption of N-Zn and Zn-O bonds under harsh conditions and that Zn-O bonds are weaker than N-Zn bonds at <i>T</i> ≥ 300 K. Broadly, we show that bond strength estimates are a reasonable indicator of Zn₂(BDC)₂(DABCO)'s performance at high temperatures and mechanical loading, and demonstrate the viability of employing AIMD simulations and physics-constrained neural networks to quantify interatomic bond strengths of hybrid organic-inorganic materials.