WilkinsonAFIRdb and related

Databases for all data related to the article: "Challenges for Kinetics Predictions via Neural Network Potentials: a Wilkinson’s catalyst case"

Each dataset was created with ASE db, and can be explored with:

import ase.db

with ase.db.connect(db_path) as db:
for row in db.select():
atoms = row.toatoms() # ASE Atoms object
data = row.data # Diverse information (energy, gradients and dipole, at DFT, xTB [and NNP or NNP(+xTB)], geometry type, reaction path network connection, ...)

Data labels: 

data['energy']: DFT energy [eV]

data['gradients']: DFT gradients [eV/A]

data['dipole']: DFT dipole [Debye]

data['xTB']['GFN2-xTB']['energy']: xTB energy [eV] (when available)

data['xTB']['GFN2-xTB']['gradients']: xTB gradients [eV/A] (when available)

data['xTB']['GFN2-xTB']['dipole']: xTB dipole [Debye] (when available)

data['E_pred']: Prediction energy [eV] (NNP, NNP(+xTB), xTB, depending on the dataset), if available

data['grad_pred']: Prediction gradients [eV/A]

data['dipole_pred']: Prediction gradients [Debye]

data['geo_type']: Type of geometry ('EQ': Equilibrium state, 'TS': Transition state, 'NODE': intermediary geometry, 'TSEQ': barrier-less TS [both path top and path endpoint])

data['EQ_id']: GRRM EQ number (sort of exploration timestamp on EQs), when available

data['TS_id']: GRRM path number (exploration timestamp on paths), when available

data['node_id']: Position in path, when available

Datasets:

WilkinsonAFIRdb.db: DFT-powered AFIR-based search data (including the single geometry with failed xTB convergence)

pureNNP_20%_dataset.zip: train/val/test data from NNP model trained on the first 20% of DFT paths explored

pureNNP_50%_dataset.zip: train/val/test data from NNP model trained on the first 50% of DFT paths explored

pureNNP_80%_dataset.zip: train/val/test data from NNP model trained on the first 80% of DFT paths explored

pureNNP_20%_localSearch.db: local NNP-powered AFIR-based search data, using NNP model trained on the first 20% of DFT paths explored

pureNNP_50%_localSearch.db: local NNP-powered AFIR-based search data, using NNP model trained on the first 50% of DFT paths explored

pureNNP_80%_localSearch.db: local NNP-powered AFIR-based search data, using NNP model trained on the first 80% of DFT paths explored

NNPxTB_20%_localSearch: local NNP-powered AFIR-based search data, using NNP(+xTB) model trained on the first 20% of DFT paths explored

NNPxTB_50%_localSearch: local NNP-powered AFIR-based search data, using NNP(+xTB) model trained on the first 50% of DFT paths explored

NNPxTB_80%_localSearch: local NNP-powered AFIR-based search data, using NNP(+xTB) model trained on the first 80% of DFT paths explored

xTB_localSearch: xTB-powered AFIR-based search data

NNPxTB_20%_globalSearch: global/full NNP-powered AFIR-based search data, using NNP(+xTB) model trained on the first 20% of DFT paths explored (EQ and TS only)

NNPxTB_50%_globalSearch: global/full NNP-powered AFIR-based search data, using NNP(+xTB) model trained on the first 50% of DFT paths explored (EQ and TS only)

Note: DFT level of theory is RωB97X-D/Def2-SVP

本数据集涵盖论文《基于神经网络势的动力学预测挑战：以威尔金森催化剂（Wilkinson’s catalyst）为例》相关的全部数据。

所有数据集均基于ASE数据库（ASE db）构建，可通过如下代码进行探索：
import ase.db

with ase.db.connect(db_path) as db:
for row in db.select():
atoms = row.toatoms() # ASE原子对象（ASE Atoms object）
data = row.data # 包含多种信息，例如密度泛函理论（DFT）、xTB[以及神经网络势（Neural Network Potential, NNP）或NNP(+xTB)]下的能量、梯度与偶极矩，几何结构类型、反应路径网络连接关系等

数据标签说明如下：
1. `data['energy']`：DFT能量，单位为电子伏特（eV）
2. `data['gradients']`：DFT梯度，单位为电子伏特每埃（eV/Å）
3. `data['dipole']`：DFT偶极矩，单位为德拜（Debye）
4. `data['xTB']['GFN2-xTB']['energy']`：GFN2-xTB方法计算的xTB能量，单位为eV（若可用）
5. `data['xTB']['GFN2-xTB']['gradients']`：GFN2-xTB方法计算的xTB梯度，单位为eV/Å（若可用）
6. `data['xTB']['GFN2-xTB']['dipole']`：GFN2-xTB方法计算的xTB偶极矩，单位为德拜（若可用）
7. `data['E_pred']`：预测能量，单位为eV（根据数据集不同，可来自NNP、NNP(+xTB)或xTB方法，若可用）
8. `data['grad_pred']`：预测梯度，单位为eV/Å
9. `data['dipole_pred']`：预测偶极矩，单位为德拜（原文此处存在笔误，原标注为"预测梯度"）
10. `data['geo_type']`：几何结构类型，可选值包括：
- `'EQ'`：平衡态（Equilibrium state）
- `'TS'`：过渡态（Transition state）
- `'NODE'`：中间几何结构（intermediary geometry）
- `'TSEQ'`：无势垒过渡态（即路径顶点与路径终点均为此类结构，barrier-less TS [both path top and path endpoint]）
11. `data['EQ_id']`：广义反应面映射（GRRM）平衡态编号，可视为平衡态的探索时间戳，若可用
12. `data['TS_id']`：GRRM路径编号，可视为反应路径的探索时间戳，若可用
13. `data['node_id']`：反应路径中的节点位置，若可用

数据集详情如下：
1. `WilkinsonAFIRdb.db`：基于DFT的自适应力诱导反应（AFIR）搜索数据集，包含xTB收敛失败的单个几何结构
2. `pureNNP_20%_dataset.zip`：基于前20%已探索DFT路径训练得到的NNP模型的训练/验证/测试数据集
3. `pureNNP_50%_dataset.zip`：基于前50%已探索DFT路径训练得到的NNP模型的训练/验证/测试数据集
4. `pureNNP_80%_dataset.zip`：基于前80%已探索DFT路径训练得到的NNP模型的训练/验证/测试数据集
5. `pureNNP_20%_localSearch.db`：基于前20%已探索DFT路径训练得到的NNP模型的局部NNP辅助AFIR搜索数据集
6. `pureNNP_50%_localSearch.db`：基于前50%已探索DFT路径训练得到的NNP模型的局部NNP辅助AFIR搜索数据集
7. `pureNNP_80%_localSearch.db`：基于前80%已探索DFT路径训练得到的NNP模型的局部NNP辅助AFIR搜索数据集
8. `NNPxTB_20%_localSearch`：基于前20%已探索DFT路径训练得到的NNP(+xTB)模型的局部NNP辅助AFIR搜索数据集
9. `NNPxTB_50%_localSearch`：基于前50%已探索DFT路径训练得到的NNP(+xTB)模型的局部NNP辅助AFIR搜索数据集
10. `NNPxTB_80%_localSearch`：基于前80%已探索DFT路径训练得到的NNP(+xTB)模型的局部NNP辅助AFIR搜索数据集
11. `xTB_localSearch`：基于xTB的AFIR搜索数据集
12. `NNPxTB_20%_globalSearch`：基于前20%已探索DFT路径训练得到的NNP(+xTB)模型的全局/全量NNP辅助AFIR搜索数据集（仅包含平衡态与过渡态结构）
13. `NNPxTB_50%_globalSearch`：基于前50%已探索DFT路径训练得到的NNP(+xTB)模型的全局/全量NNP辅助AFIR搜索数据集（仅包含平衡态与过渡态结构）

注：本次计算所采用的DFT理论级别为RωB97X-D/Def2-SVP