PyXtal as a library

While the PyXtal can be used in the command mode, it can become much more powerful with Python scripting. Here we describe the basic functionality of PyXtal as a Python Library. This tutorial aims to cover the following contents:

  • Built-in PyXtal tools
  • Crystal structure generation
  • Crystal structure manipulation

Available Tools in PyXtal

PyXtal includes the following functions:


The Group package makes working with symmetry groups simple. Useful information can be accessed directly as follows:

>>> from pyxtal.symmetry import Group
>>> g = Group(45)
>>> g
-- Space group # 45 --
  8c        site symm: 1
  4b        site symm: ..2
  4a        site symm: ..2
>>> g.chiral   # check if the space group is enantiomorphic
>>> g.inversion #check if it has inversion symmetry
>>> g.polar #check if it is polar

It is important to note that one space group may have multiple settings (see the Settings page for details). To avoid the ambiguity, Hall introduced the explicit-origin space group notation. Following the Hall notation, there exist 530 Concise space groups. The full list is available online. In PyXtal, we also the initialization of space group according to Hall number. Below shows an example to create the Group object of Fd-3m (227) with the choice 1 of origin.

>>> g = Group(525, use_hall=True)
>>> g.symbol
'F d -3 m:1'
>>> g[-1]
Wyckoff position 8a in space group 227 with site symmetry -4 33 mm
1/4, 1/4, 1/4
1/2, 0, 1/2

For a comparison, we also show Fd-3m (227) in the standard setting of with the choice 2 of origin. These two notations only differ in which symmetry point is placed at (0,0,0).

>>> g = Group(526, use_hall=True)
>>> g.symbol
'F d -3 m:2'
>>> g[-1]
Wyckoff position 8a in space group 227 with site symmetry -4 33 mm
1/8, 1/8, 1/8
3/8, 7/8, 3/8

If one wants to follow the spglib style to initialize the Group object, the following way should work,

>>> g = Group(227, style='spglib')
>>> g.hall_number

Layer, rod, and point groups can be accessed by passing the parameter `dim=2, dim=1, or dim=0, respectively.

>>> Group(5, dim=2)
-- Layer group # 5 --
  2a        site symm: 1
>>> Group(5, dim=1)
-- Rod group # 5 --
  2a        site symm: 1
>>> Group(5, dim=0)
-- Point group 5 --
  4d        site symm: 1
  2c        site symm: m . .
  2b        site symm: 2 . .
  1a        site symm: 2/m . .

A Group instance contains the Wyckoff positions, site symmetry, and generators for the group. In addition, the Group class stores the lattice type (lattice_type), international number (number), symbol (symbol), and the periodic boundary conditions (PBC). Each group is divided into Wyckoff positions, which are sets of points which possess some subset of the complete group symmetry.


A Wyckoff position is typically denoted with a number-letter combination, depending on its multiplicity. For example, for space group Iba2 (45), the general Wyckoff position is called 8c. This means the position has a multiplicity of 8. The letters a and b are used by special Wyckoff positions. Note that the name convention is different for point groups; a point group may have the special Wyckoff position 1o, which corresponds to the point (0,0,0). This is in contrast to the default name 1a. Each Wyckoff position is further separated into individual operations ('-x,-y,z', '1,1/2,z+1/2', etc.).

When a Group is defined, its Wyckoff_position can be accessed with either a numerical index or letter.

>>> g[0]
Wyckoff position 8c in space group 45 with site symmetry 1
x, y, z
-x, -y, z
x+1, -y+1, z+1/2
-x+1, y+1, z+1/2
>>> g['b']
Wyckoff position 4b in space group 45 with site symmetry ..2
0, 1/2, z
1, 1/2, z+1/2

As displayed in the example above, the Wyckoff position 4b has site symmetry ..2. In this example, . denotes no symmetry about the x and y axes, and 2 denotes a 2-fold rotation about the z axis in Hermann-Mauguin notation. In each WP, the symmetry operations are stored as SymmOp objects. These symmetry operations can be applied to 3d vectors using op.operate, or can be composed together via multiplication: op3 = op1 * op2. Each SymmOp consists of a rotation matrix (op.rotation_matrix) and a translation vector (op.translation_vector), and is represented by a :math: 4 times 4 affine matrix (op.affine_matrix).

Alternatively, the WP can be initialized by itself.

>>> from pyxtal.symmetry import Wyckoff_position as wp
>>> wp.from_group_and_index(19, 0)
Wyckoff position 4a in space group 19 with site symmetry 1
x, y, z
-x+1/2, -y, z+1/2
-x, y+1/2, -z+1/2
x+1/2, -y+1/2, -z


There are three options for defining molecules within Pyxtal. First, you need to import the pyxtal_molecule class,

from pyxtal.molecule import pyxtal_molecule
  1. From a pre-defined string for the chemical composition
mol = pyxtal_molecule('H2O')

The list of supported molecules are accessible via:

>>> pyxtal_molecule.list_molecules()
>>> ['C60', 'Glycine-z', 'xxvi', 'xxv', 'BIPHEN', 'ANULEN',
'H2O', 'CH4', 'NH3', 'benzene', 'naphthalene', 'anthracene',
'tetracene', 'Pentacene', 'coumarin', 'resorcinol', 'benzamide',
'aspirin', 'ddt', 'lindane', 'Glycine', 'Glucose', 'ROY', 'LEFCIK',
  1. From a Molecule object.
from pymatgen.core import Molecule

Water molecule
O          0.00000        0.00000        0.11779
H          0.00000        0.75545       -0.47116
H          0.00000       -0.75545       -0.47116

m = Molecule.from_str(xyz, fmt='xyz')
mol = pyxtal_molecule(m)

# Alternatively, one can load a xyz molecule file.
# It will be converted to pymatgen.molecule and then passed to pyxtal.
mol = pyxtal_molecule('')

3) a smile string representing the molecule. For example, C1=CC=CC=C1.smi means a benzene molecule. Note that the .smi suffix must be included to indicate that this is a smile string. In this case, RDKit must be installed to use this function.. One can install RDKit by simply typing

$ conda install -c conda-forge rdkit==2021.09.2.

Note that the current code is designed for version no later than 2021.09.2.

mol = pyxtal_molecule('CC(=O)NC1=CC=CC=C1C(=O)N.smi')

After the molecule is defined, its point group will also be parsed:

mol = pyxtal_molecule('H2O')
-- Pointgroup --# 7 (C2v)--
4d  site symm: 1
2c  site symm: m . .
2b  site symm: m . .
1a  site symm: mm2 . .


It is possible to supply your own unit cell lattice for a random crystal, via the Lattice class. You can define a lattice using either a :math: 3 times 3 matrix, or 6 cell parameters:

from pyxtal.lattice import Lattice
l1 = Lattice.from_matrix([[4.08,0,0],[0,9.13,0],[0,0,5.50]])
l2 = Lattice.from_para(4.08, 9.13, 5.50, 90, 90, 90)

Here, both l1 and l2 describe the same lattice. In this case, it is an orthorhombic cell with lengths 4.08, 9.13, and 5.50 \(\\AA\), which is the unit cell for common water ice. The lattice parameters are, in order: (a, b, c, \(\alpha, \beta, \gamma\)). a, b, and c are the lengths of the lattice vectors; \(\alpha, \beta, \gamma\) are the angles in degrees between these vectors.


When generating random crystals, PyXtal performs inter-atomic distances checks to make sure the atoms are not too close together. By default, the covalent radius is used as a basis. However, the user may also define their own criteria using the Tol_matrix class. To do this, initialize a Tol_matrix object using one of the built-in methods.

from pyxtal.tolerance import Tol_matrix
tol_m_1 = Tol_matrix(prototype="molecular", factor=2.0)
tol_m_2 = Tol_matrix.from_radii(some_custom_list_of_atomic_radii)
tol_m_3 = Tol_matrix.from_matrix(some_custom_2D_tolerance_matrix)

Crystal structure generation

PyXtal allows one to generate the crystal from either the existing structure or from the scratch. First, One can always load an existing crystal from a given file path. More importantly, PyXtal can generate the trial structure according to the customized factors such as space group, cell parameters, partial occupation. It also supports on handling different systems from atomic to molecular, and from 1D to 3D.

Loading the existing structures

Assuming there is a file in either cif or VASP POSCAR format, one can just load the structure by the from_seed function.

from pyxtal import pyxtal
my_crystal = pyxtal()

my_crystal.from_seed(seed=struc_file, style='pyxtal')
my_crystal.from_seed(seed=struc_file, style='spglib')

Note that the style flag allows one to represent the structure in different space group settings.

For the molecular crystal, the molecular information must be provided as a list (see the molecule section for more details).

from pyxtal import pyxtal
my_crystal = pyxtal(molecular=True)
my_crystal.from_seed(seed=struc_file, molecule=['aspirin'])

In addition to the existing files in either cif or VASP POSCAR, pyxtal also provides the interface with Pymatgen and ASE, which support a variety of structure formats. Below we show a few working examples.

from pyxtal import pyxtal
c = pyxtal()

# load the structure from ase
from import read
ase_atoms = read('1.cif', format='cif')

# load the structure from pymatgen
from pymatgen.core import Structure
pmg = read('1.cif', format='cif')

Random 3D Atomic Crystals

PyXtal allows the user to generate random crystal structures with given symmetry constraints. There are several parameters which can be specified, but only three are necessary:

  • the symmetry group,
  • the types of atoms,
  • the number of each atom in the primitive cell

Here is a simple example of a 3D carbon crystal:

from pyxtal import pyxtal
my_crystal = pyxtal()
my_crystal.from_random(3, 225, ['C'], [12])

This would create a crystal structure with 3D structure with space group 225, 12 carbon atoms in the conventional cell. For stoichiometry with more than one type of atom, replace [C] with a list of atomic symbols, and replace [12] with a list of numbers. For example,

>>> my_crystal = pyxtal()
>>> my_crystal.from_random(3, 99, ['Ba','Ti','O'], [1,1,3])
>>> my_crystal
------Random Crystal------
Composition: Ba1 Ti1 O3
Dimension: 3
Group: P4mm (99)
Volume factor: 1.0
tetragonal lattice:   5.1029   5.1029   4.3018  90.0000  90.0000  90.0000
Wyckoff sites:
    Ba @ [0.5000 0.5000 0.3612], Wyckoff letter:  1b, Site symmetry: 4 m m
    Ti @ [0.5000 0.5000 0.8701], Wyckoff letter:  1b, Site symmetry: 4 m m
     O @ [0.5000 0.0000 0.0823], Wyckoff letter:  2c, Site symmetry: 2 mm .
     O @ [0.5000 0.5000 0.8177], Wyckoff letter:  1b, Site symmetry: 4 m m

would create a random \(BaTiO_3\) crystal. If the generation is successful, the value of my_crystal.valid will be set to True; otherwise, it will be False.

Random 3D molecular crystals

3D Molecular crystals are generated in the same way as atomic crystals, but atomic species are replaced with (rigid) molecules. The following script would give a crystal with space group 36, 4 molecules in the conventional unit cell.

my_crystal = pyxtal(molecular=True)
my_crystal.from_random(3, 36, ['H2O'], [4])

------Random Molecular Crystal------
Dimension: 3
Group: Cmc21
Volume factor: 1.0
orthorhombic lattice:   5.6448   6.3389   4.4262  90.0000  90.0000  90.0000
Wyckoff sites:
    H2 O1 @ [ 0.000  0.596  0.986]  Wyckoff letter:  4a, Site symmetry m.. ==> Rotvec: -0.343  0.000  0.000

For molecular crystals, it is possible that a structure is better represented in a non-standard setting. PyXtal supports the generation of crystals from a non-standard setting (as defined by the Hall number). Below compares how to generate the crystals of \(P2_1/c\) and \(P2_1/n\), which are both in space group 14.

>>> from pyxtal import pyxtal
>>> c1 = pyxtal(molecular=True)
>>> c1.from_random(3, 81, ["aspirin"], use_hall=True)
>>> c1
------Crystal from random------
Dimension: 3
Composition: [aspirin]4
Group: P 1 21/c 1 (14)
12.6259,  15.1971,  12.3168,  90.0000,  84.2525,  90.0000, monoclinic
Wyckoff sites:
    H8C9O4       @ [ 0.6281  0.9928  0.7032]  WP [4e] Site [1] Euler [  57.4  -46.9   89.8]

>>> c1.from_random(3, 82, ["aspirin"], use_hall=True)
>>> c1
------Crystal from random------
Dimension: 3
Composition: [aspirin]4
Group: P 1 21/n 1 (14)
16.4395,  16.5499,   9.4357,  90.0000, 113.6587,  90.0000, monoclinic
Wyckoff sites:
    H8C9O4       @ [ 0.0181  0.6252  0.5789]  WP [4e] Site [1] Euler [-179.0   46.1  -63.9]

Random sub-periodic crystals

PyXtal can also generate sub-periodic crystals. For example,

my_crystal = pyxtal()
my_crystal.from_random(2, 20, ['C'], [4], thickness=2.0)

would generate a 2d crystal with

  • layer group P2_122 (20),
  • 4 carbon atoms in the conventional cell,
  • a thickness of 2.0 \(\\AA\).

The crystal will be periodic in two directions instead of three. PyXtal adds 10 \(\\AA\) of vacuum on the z axis (which is non-periodic). Note that the layer group number is different from the space group number, and ranges between 1 and 80. By default, PyXtal will automatically generate a value for the thickness of the unit cell, based on the volume. By specifying thickness value, you override this behavior. So, if you are testing over a range of volume factors, consider how the shape of the unit cell will be affected, and change the thickness accordingly. Alternatively, you may supply a custom Lattice object.

You can generate 1D crystals using Rod groups (between 1 and 75) and atomic clusters with point group symmetry.

1d = pyxtal()
1d.from_random(1, 20, ['C'], [4])

0d= pyxtal()
0d.from_random(0, 'Ih', ['C'], [60])

The point group may be specified either by a number (only for the crystallographic point groups), or by a symbol (see the Settings page).

2D and 1D molecular crystals are also supported.

my_crystal = pyxtal()
my_crystal.from_random(2, 20, ['H2O'], [4])
my_crystal.from_random(1, 20, ['H2O'], [4])

Crystal structure manipulation

After the crystal is built, PyXtal allows one to manipulate the structure in different ways. The following script illustrate some useful functions.

# create a random crystal
c = pyxtal()
c.from_random(3, 227, ['C'], [8])

------Crystal from random------
Dimension: 3
Composition: C8
Group: F d -3 m:2 (227)
4.9107,   4.9107,   4.9107,  90.0000,  90.0000,  90.0000, cubic
Wyckoff sites:
    C @ [ 0.1250  0.1250  0.1250], WP [8a] Site [-433mm]

# get a subgroup representation
------Crystal from subgroup------
Dimension: 3
Composition: C8
Group: I 41/a m d:2 (141)
3.4724,   3.4724,   4.9667,  90.0000,  90.0000,  90.0000, tetragonal
Wyckoff sites:
    C @ [ 0.0000  0.7500  0.1250], WP [4a] Site [-4mm2]

# compute the pxrd
>>> c.get_XRD()
  2theta     d_hkl     hkl       Intensity  Multi
  31.556     2.835   [ 1  1  1]   100.00        8
  52.723     1.736   [ 2  2  0]    42.05       12
  62.755     1.481   [ 3  1  1]    21.09       24
  77.799     1.228   [ 4  0  0]     5.08        6
  86.361     1.127   [ 3  3  1]     7.87       24
 100.543     1.002   [ 4  2  2]    12.92       24
 109.320     0.945   [ 5  1  1]     8.55       24
 125.261     0.868   [ 4  4  0]     7.45       12
 136.483     0.830   [ 5  3  1]    18.32       48
 166.319     0.776   [ 6  2  0]    58.30       24

In addition, the structure can be exported to a variety of formats for further analysis and process.

from pyxtal import pyxtal
c = pyxtal()
c.from_random(3, 225, ['C'], [16])

# export the structure to pymatgen or ase.Atoms object.
pmg_struc = c.to_pymatgen()
ase_struc = c.to_ase()

# ase.Atoms object supports a lot of methods for structural manipulation
ase_struc *= 2             # create 2*2*2 supercell
ase_struc *= [1, 2, 2]     # create 1*2*2 supercell

# Export the structure into different formats
ase_struc.write('1.vasp', format='vasp', vasp5=True, direct=True)
ase_struc.write('', format='extxyz')

For the molecular crystals, the atomic order will automatically adjusted when converting when the structure is converted to ASE Atoms object. If you want to keep the original order, just set resort=False when calling the to_ase() function.

my_crystal = pyxtal()
my_crystal.from_random(3, 36, ['H2O'], [4], 1.0)
xtal = my_crystal.to_ase(resort=False)

Atoms(symbols='OH2OH2OH2OH2', pbc=True, cell=[[6.503138824544265, 0.0, 0.0], [3.0183112928813903e-16, 4.929276416649856, 0.0], [3.025303230945897e-16, 3.025303230945897e-16, 4.940695118057273]])

ordered_xtal = my_crystal.to_ase()
Atoms(symbols='H8O4', pbc=True, cell=[[6.503138824544265, 0.0, 0.0], [3.0183112928813903e-16, 4.929276416649856, 0.0], [3.025303230945897e-16, 3.025303230945897e-16, 4.940695118057273]])

Advanced examples in random structure generation

In addition to the required parameters, the user can provide additional constraints.

Constraints on lattice and sites

Sometimes, it is convenient to generate the crystal with partial information. Below shows how to create a \(Al_2SiO_5\) crystal with a pre-assigned unit cell and sites on 8Al + 4Si + 4O, and random coordinates on the 16 remaining O atoms.

from pyxtal.lattice import Lattice
cell = Lattice.from_para(7.8758, 7.9794, 5.6139, 90, 90, 90, ltype='orthorhombic')
spg = 58
elements = ['Al', 'Si', 'O']
composition = [8, 4, 20]

sites = [{"4e": [0.0000, 0.0000, 0.2418],
          "4g": [0.1294, 0.6392, 0.0000],
         {"4g": [0.2458, 0.2522, 0.0000]},
         {"4g": [0.4241, 0.3636, 0.0000]}, #partial information on O sites

s = pyxtal()
s.from_random(3, spg, elements, composition, lattice=cell, sites=sites)

------Crystal from random------
Dimension: 3
Composition: O20Si4Al8
Group: Pnnm (58)
  7.8758,   7.9794,   5.6139,  90.0000,  90.0000,  90.0000, orthorhombic
Wyckoff sites:
    Al @ [ 0.0000  0.0000  0.2418], WP [4e] Site [..2]
    Al @ [ 0.1294  0.6392  0.0000], WP [4g] Site [..m]
    Si @ [ 0.2458  0.2522  0.0000], WP [4g] Site [..m]
     O @ [ 0.4241  0.3636  0.0000], WP [4g] Site [..m]
     O @ [ 0.5538  0.2648  0.0000], WP [4g] Site [..m]
     O @ [ 0.0000  0.5000  0.6057], WP [4f] Site [..2]
     O @ [ 0.8809  0.5970  0.0786], WP [8h] Site [1]

Similarly, PyXtal allows the user to pre-assign the partial information (e.g., lattice, Wyckoff sites) before generating the crystals. A list of scripts is shown below.

s = pyxtal()
# Generatation with minimum input
s.from_random(from_random(3, 14, ['aspirin'], [4])

# Add Lattice constraints
from pyxtal.lattice import Lattice
lat = Lattice.from_para(11.43, 6.49, 11.19, 90, 83.31, 90, ltype='monoclinic')
s.from_random(3, 14, ['aspirin'], [4], lattice=lat)

# Add sites constraints
sites = [{"4e": [0.77, 0.57, 0.53]}]
s.from_random(3, 14, ['aspirin'], [4], lattice=lat, sites=sites)

# Crystal with 2 water molecules occupying two special wyckoff sites
# This requires that the molecule is compatible with the site symmetry, be cautious!
s.from_random(3, 36, ["H2O"], [8], sites=[["4a", "4a"]])

Random molecular crystal without calling pyxtal_molecule

If you just want to generate a random molecular crystal, Pyxtal will automatically interpret the strings. Therefore, it is not necessary to call the pyxtal_molecule class. See a short example below.

from pyxtal import pyxtal
c1 = pyxtal(molecular=True)
c1.from_random(3, 14, ['CC(=O)NC1=CC=CC=C1C(=O)N.smi'], [4])

Random molecular crystal with constraints on torsion

Using the smile string, one can specify the desired torsions

from pyxtal import pyxtal

c1 = pyxtal(molecular=True)
c1.from_random(3, 14, ['CC(=O)NC1=CC=CC=C1C(=O)N.smi'], [4], torsions=[[-60.2, 1.7, 126.5]])
print("Torsions", c1.mol_sites[0].encode()[-4:-1])
------Crystal from random------
Dimension: 3
Composition: [CC(=O)NC1=CC=CC=C1C(=O)N]4
Group: P21/c (14)
monoclinic lattice:  19.2246  13.2842  10.1448  90.0000 113.3669  90.0000
Wyckoff sites:
        H10C9N2O2 @ [ 0.2497  0.4534  0.9597]  WP:  4e, Site symmetry 1 ==> Euler: -66.31  25.98 -37.99
Torsions [-60.19971274864328, 1.6999253045986045, 126.50111998425088]

Symmetry Compatibility in Molecular Crystals

For the molecules with high point group symmetry, it is possible that the molecule can occupy the special Wyckoff site. Different from other codes, PyXtal offers an internal function to check if the molecular symmetry is compatible with the Wyckoff site symmetry. Below is a short example to illustrate the function.

from pyxtal.symmetry import Group
from pyxtal.molecule import pyxtal_molecule

mol = pyxtal_molecule('H2O')
sgs = [14, 36, 63]

for sg in sgs:
    spg = Group(sg)
    for wp in spg.Wyckoff_positions:
        if len(mol.get_orientations_in_wp(wp)) > 0:

If you run the above script, it is expected to return all the possible Wyckoff sites that can host the H2O molecule.

Wyckoff position 4e in space group 14 with site symmetry 1
Wyckoff position 8b in space group 36 with site symmetry 1
Wyckoff position 4a in space group 36 with site symmetry m..
Wyckoff position 16h in space group 63 with site symmetry 1
Wyckoff position 8g in space group 63 with site symmetry ..m
Wyckoff position 8f in space group 63 with site symmetry m..
Wyckoff position 8e in space group 63 with site symmetry 2..
Wyckoff position 4c in space group 63 with site symmetry m2m

1D Representation (Experimental)

For the molecular crystal, PyXtal also provides a representation class to handle the conversion between Pyxtal and its 1D representation. With this module, one can represent the crystal into a 1D array.

from pyxtal import pyxtal

c1 = pyxtal(molecular=True)
c1.from_seed('pyxtal/database/cifs/aspirin.cif', ['CC(=O)OC1=CC=CC=C1C(=O)O.smi'])
rep = c1.get_1D_representation()
81 11.23  6.54 11.23  95.9 1 0.23 0.59 0.03   44.1  -25.2   32.5   82.9    2.8 -178.3 1

In the 1D string, the data is organized as follows

  • Hall number (1-530)
  • cell parameter: a, b, c, alpha, beta, gamma
  • molecular site: center coordinates + orientation + torsions + inversion

Alternatively, one can read the structure from the 1D representation and smile string

from pyxtal.representation import representation
rep1 = representation(rep.x, ['CC(=O)OC1=CC=CC=C1C(=O)O'])
xtal = rep1.to_pyxtal()
------Crystal from 1D rep.------
Dimension: 3
Composition: [CC(=O)OC1=CC=CC=C1C(=O)O]4
Group: P 1 21/c 1 (14)
11.2330,   6.5440,  11.2310,  90.0000,  95.8900,  90.0000, monoclinic
Wyckoff sites:
    H8C9O4       @ [ 0.2252  0.5852  0.0308]  WP [4e] Site [1] Euler [  44.1  -25.2   32.5]