- 學位論文
有機金屬框架儲鋰性能之多尺度原子模擬
Multiscale Atomistic Simulations of Li ion Storage in Metal-Organic Frameworks
摘要
有機金屬框架（Metal-organic frameworks，MOFs）為金屬原子或原子簇與有機化合物配位組成之結構，是目前新形態混合有機與無機材料中研究最熱門的領域。自從90年代，科學家們成功的合成出孔洞有固定排序且穩定結構的MOFs材料之後，不同種類、孔隙率、孔尺寸大小的MOFs材料不斷出現。至今，由於MOFs材料具有孔徑尺寸可調、功能性強和比表面積大、具有多樣性金屬與官能團等特性，MOFs材料已經被廣泛運用在各個領域，例如：氣體儲存、氣體分離、催化劑、傳感器及螢光應用等。 本研究將探討兩種不同的Pb-MOFs材料之儲鋰性能，分別是以硝酸鉛與均苯三甲酸合成的(1-羧基-3,5-羧酸根苯)二水合鉛(II)以及由乙酸鉛與均苯三甲酸合成的二(1,3,5-羧酸根苯)一水化合鉛(II)兩種材料，由於兩者擁有高表面積與孔隙率大等性質，因此皆具有當作鋰離子電池負極材料的潛力。 本文首先運用第一原理密度泛函理論(DFT)探討是否去除結構中之結晶水分子，接著計算的不同數量的鋰離子在MOFs中的吸附機制與穩定吸附位置，並與實驗上的結果做對照。以現今分子模擬來說，第一原理計算雖然能夠精確地描述原子間的相互作用力與能量，但需要花費非常高的運算資源及時間，系統模擬的尺度也有所受限。在先前的研究中，已成功證實了機器學習訓練搭配適當的descriptor能夠得到精確描述原子間受力與能量的人工神經網路勢能，其計算花費遠小於DFT方法，又能夠推廣到較大的模擬系統。因此本研究以前述小尺度MOFs吸附鋰離子的DFT計算結果當作訓練集，訓練出能夠描述該材料的勢能模型。隨後，利用此勢能利用分子動力學與巨正則系綜蒙地卡羅方法，可以在較大的尺度下，探討鋰離子在該材料中偏好吸附的位置並與DFT的計算結果比較。透過多尺度的模擬方法，觀察Pb-MOFs吸附鋰離子時儲存機制並計算其理論電容量，因此利用分子模擬方法可以提供有用的資訊使實驗團隊在研發新穎鋰離子電池材料更加有效率。
並列摘要
Metal-organic frameworks (MOFs) are a kind of compounds consisting of clusters or metal ions coordinated to organic ligands structures and have been an active research area since its first successful synthesis in the 1990s. Because of their tunable porosity, high surface area, as well as the diversity in the combination of metal and organic functional groups, MOFs have been utilized over a wide range of applications such as luminescence, gas separation and separation, heterogeneous catalysis, and sensors. In this thesis, we investigated the Li ion storage mechanisms of two Pb-MOFs for battery applications, namely, the Pb2(1,3,5-HBTC)(H2O)2 and Pb_3(1,3,5-BTC)2(H2O). We first employed the density functional theory (DFT) calculations to explore the role of water molecules for the structural stability of the MOFs. Then, we examined the Li adsorption sites and computed the respective adsorption energies as the function of the number of inserted Li ions. The capacities and volume expansion during Li insertion from DFT calculations are in good agreements with experiments; hence, our DFT calculations provided atomistic insights into the Li adsorption processes in both MOFs, which are extremely difficult to be extracted from experiments. Despite we can use DFT calculations to gain insights into the Li adsorption behaviors in both MOFs with good agreements with experiments; however, the DFT calculations are computationally expensive, thereby imposing limitations to both the spatial and temporal length scales of Li insertion simulations. With the recent progress in machine learning, it is possible to train a machine-learning-enabled, artificial neural network (ANN) energy model that can predict the potential energy and atomic forces of MOFs orders of magnitude faster than DFT calculations, while retaining high fidelity to DFT calculations. The trained ANN potential model can accurately describe atomic interactions of MOFs, and we can perform molecular dynamics simulations as well as grand canonical Monte Carlo (GCMC) simulations to explore the Li adsorption behaviors with system size beyond the reach of DFT calculations. We demonstrate that using ANN potential model in conjunction with GCMC simulations we could simulate the adsorption processes with much larger system and shorter time than DFT calculations, and the capacity from GCMC simulations are once again in good agreement with experiments. Therefore, by using machine-learning-enabled multiscale atomistic simulation approach it is possible to examine the Li storage mechanism in MOFs and compute the theoretical capacity, thereby helping experimental teams develop advanced high-capacity lithium-ion batteries.