import lmdb
import pickle
import gzip

lmdb_file = '/path/to/lmdb/dir/hokusai2017.lmdb'

with lmdb.open(lmdb_file, readonly=True, subdir=False) as env:
    with env.begin() as txn:
        val = pickle.loads(gzip.decompress((txn.get(b'000000984'))))
    
pm3_val = val['pm3']
hf_val = val['hf']
dft1st_val = val['DFT_1st']
dft2nd_val = val['DFT_2nd']

for step in dft1st_val:
    
    # coords & grad is a list of dictionaries that stores the relevant information of each atom
    # energy is a scalar representing the energy for that conformer
    
    coords = step['coordinates']
    energy = step['energy']
    grad = step['gradient']
    
    for atom in coords:
        # access atom's attributes
        element = atom['atom']
        x = atom['x']
        y = atom['y']
        z = atom['z']
        
    for atom in grad:
        # access atom's attributes
        element = atom['atom']
        dx = atom['dx']
        dy = atom['dy']
        dz = atom['dz']