A database for atoms

ASE has its own database that can be used for storing and retrieving atoms and associated data in a compact and convenient way.

There are currently five back-ends:

JSON:

Simple human-readable text file with a .json extension.

SQLite3:

Self-contained, server-less, zero-configuration database. Lives in a file with a .db extension.

PostgreSQL:

Server based database.

MySQL:

Server based database.

MariaDB:

Server based database.

The JSON and SQLite3 back-ends work “out of the box”, whereas PostgreSQL, MySQL and MariaDB requires a server (See Running a PostgreSQL server or Running a MySQL server).

There is a command-line tool called ase db that can be used to query and manipulate databases and also a Python interface.

What’s in the database?

Every row in the database contains:

• all the information stored in the Atoms object (positions, atomic numbers, …)

• calculator name and parameters (if a calculator is present)

• already calculated properties such as energy and forces (if a calculator is present)

• key-value pairs (for finding the calculation again)

• an integer ID (unique for each database) starting with 1 and always increasing for each new row

• a unique ID which is a 128 bit random number which should be globally unique (at least in the lifetime of our universe)

• constraints (if present)

• user-name

• creation and modification time

ase db

The ase db command-line tool can be used to query databases and for manipulating key-value pairs. Try:

$ ase db --help

Example: Show all rows of SQLite database abc.db:

$ ase db abc.db
id|age|user |formula|calculator|energy|natoms| fmax|pbc|charge| mass
 1| 1s|jensj|H2     |emt       | 1.419|     2|9.803|FFF| 0.000|2.016
 2| 1s|jensj|H2     |emt       | 1.071|     2|0.000|FFF| 0.000|2.016
 3| 1s|jensj|H      |emt       | 3.210|     1|0.000|FFF| 0.000|1.008
Rows: 3
Keys: relaxed

Show all details for a single row:

$ ase db abc.db relaxed=1 -l
H2:
Unit cell in Ang:
axis|periodic|          x|          y|          z|    length|     angle
   1|      no|      0.000|      0.000|      0.000|     0.000|    90.000
   2|      no|      0.000|      0.000|      0.000|     0.000|    90.000
   3|      no|      0.000|      0.000|      0.000|     0.000|    90.000

Data: abc

Key     | Description | Value
id      |             | 2
ctime   | Age         | 1.586 seconds
fmax    |             | 0.000
mass    |             | 2.016
relaxed |             | True
energy  |             | 1.071

See also

• Command line tool

Querying

Here are some example query strings:

Cu	contains copper
H<3	less than 3 hydrogen atoms
Cu,H<3	contains copper and has less than 3 hydrogen atoms
H2O	At least two hydrogens and at least one oxygen
formula=H2O	Exactly two hydrogens and one oxygen
v3	has ‘v3’ key
abc=bla-bla	has key ‘abc’ with value ‘bla-bla’
v3,abc=bla-bla	both of the above
calculator=nwchem	calculations done with NWChem
2.2<bandgap<3.0	‘bandgap’ key has value between 2.2 and 3.0
natoms>=10	10 or more atoms
id=2345	specific id
age<1h	not older than 1 hour
age>1y	older than 1 year
pbc=TTT	Periodic boundary conditions along all three axes
pbc=TTF	Periodic boundary conditions along the first two axes (F=False, T=True)

These names are special:

id	integer identifier
natoms	number of atoms
pbc	Periodic boundary conditions
formula	formula
energy	potential energy
charge	total charge
magmom	total magnetic moment
calculator	name of calculator
user	who did it
age	age of calculation (use s, m, h, d, w, M and y for second, minute, hour, day, week, month and year respectively)

Integration with other parts of ASE

ASE’s ase.io.read() function can also read directly from databases:

>>> from ase.io import read
>>> a = read('abc.db@42')
>>> a = read('abc.db@id=42')  # same thing
>>> b = read('abc.db@v3,abc=H')

Also the ASE’s GUI program can read from databases using the same syntax.

Browse database with your web-browser

You can use your web-browser to look at and query databases like this:

$ ase db abc.db -w
$ firefox http://0.0.0.0:5000/

Click individual rows to see details. See the CMR web-page for an example of how this works.

Python Interface

First, we connect() to the database:

>>> from ase.db import connect
>>> db = connect('abc.db')

or

>>> import ase.db
>>> db = ase.db.connect('abc.db')

Let’s do a calculation for a hydrogen molecule and write some results to a database:

>>> from ase import Atoms
>>> from ase.calculators.emt import EMT
>>> h2 = Atoms('H2', [(0, 0, 0), (0, 0, 0.7)])
>>> h2.calc = EMT()
>>> h2.get_forces()
array([[ 0.   ,  0.   , -9.803],
       [ 0.   ,  0.   ,  9.803]])

Write a row to the database with a key-value pair ('relaxed', False):

>>> db.write(h2, relaxed=False)
1

The write() method returns an integer id.

Do one more calculation and write results:

>>> from ase.optimize import BFGS
>>> BFGS(h2).run(fmax=0.01)  
BFGS:   0  12:49:25        1.419427       9.8029
BFGS:   1  12:49:25        1.070582       0.0853
BFGS:   2  12:49:25        1.070544       0.0236
BFGS:   3  12:49:25        1.070541       0.0001
>>> db.write(h2, relaxed=True)
2

Loop over selected rows using the select() method:

>>> for row in db.select(relaxed=True):
...     print(row.forces[0, 2], row.relaxed)  
-9.8029057329 False
-9.2526347333e-05 True

The select() method will generate Row objects that one can loop over.

Write the energy of an isolated hydrogen atom to the database:

>>> h = Atoms('H')
>>> h.calc = EMT()
>>> h.get_potential_energy()
3.21
>>> db.write(h)
3

Select a single row with the get() method:

>>> row = db.get(relaxed=1, calculator='emt')
>>> for key in row:
...    print('{0:22}: {1}'.format(key, row[key]))  
pbc                   : [False False False]
relaxed               : True
calculator_parameters : {}
user                  : jensj
mtime                 : 15.3439399027
calculator            : emt
ctime                 : 15.3439399027
positions             : [[ ... ]]
id                    : 2
cell                  : [[ 0.  0.  0.] [ 0.  0.  0.] [ 0.  0.  0.]]
forces                : [[ ... ]]
energy                : 1.07054126233
unique_id             : bce90ff3ea7661690b54f9794c1d7ef6
numbers               : [1 1]

Calculate the atomization energy and update() a row in the database:

>>> e2 = row.energy
>>> e1 = db.get(H=1).energy
>>> ae = 2 * e1 - e2
>>> print(ae)
5.34945873767
>>> id = db.get(relaxed=1).id
>>> db.update(id, atomization_energy=ae)
1

Delete a single row:

>>> del db[db.get(relaxed=0).id]

or use the delete() method to delete several rows.

Description of a row

The first 9 keys (from “id” to “positions”) are always present — the rest may be there:

key	description	datatype	shape
id	Local database id	int
unique_id	Globally unique hexadecimal id	str
ctime	Creation time	float
mtime	Modification time	float
user	User name	str
numbers	Atomic numbers	int	(N,)
pbc	Periodic boundary condition flags	bool	(3,)
cell	Unit cell	float	(3, 3)
positions	Atomic positions	float	(N, 3)
initial_magmoms	Initial atomic magnetic moments	float	(N,)
initial_charges	Initial atomic charges	float	(N,)
masses	Atomic masses	float	(N,)
tags	Tags	int	(N,)
momenta	Atomic momenta	float	(N, 3)
constraints	Constraints	list of dict
energy	Total energy	float
forces	Atomic forces	float	(N, 3)
stress	Stress tensor	float	(6,)
dipole	Electrical dipole	float	(3,)
charges	Atomic charges	float	(N,)
magmom	Magnetic moment	float
magmoms	Atomic magnetic moments	float	(N,)
calculator	Calculator name	str
calculator_parameters	Calculator parameters	dict

Extracting Atoms objects from the database

If you want an Atoms object insted of an AtomsRow object, you should use the get_atoms() method:

>>> h2 = db.get_atoms(H=2)

Add additional data

When you write a row to a database using the write() method, you can add key-value pairs where the values can be strings, floating point numbers, integers and booleans:

>>> db.write(atoms, functional='LDA', distance=7.2)

More complicated data can be written like this:

>>> db.write(atoms, ..., data={'parents': [7, 34, 14], 'stuff': ...})

and accessed like this:

>>> row = db.get(...)
>>> row.data.parents
[7, 34, 14]

Row objects

There are three ways to get at the columns of a row:

1. as attributes (row.key)

2. indexing (row['key'])

3. the get() method (row.get('key'))

The first two will fail if there is no key column whereas the last will just return None in that case. Use row.get('key', ...) to use another default value.

class ase.db.row.AtomsRow(dct)

get(key, default=None)

Return value of key if present or default if not.

property key_value_pairs

Return dict of key-value pairs.

count_atoms()

Count atoms.

Return dict mapping chemical symbol strings to number of atoms.

property constraints

List of constraints.

property data

Data dict.

property natoms

Number of atoms.

property formula

Chemical formula string.

property symbols

List of chemical symbols.

property fmax

Maximum atomic force.

property constrained_forces

Forces after applying constraints.

property smax

Maximum stress tensor component.

property mass

Total mass.

property volume

Volume of unit cell.

property charge

Total charge.

toatoms(add_additional_information=False)

Create Atoms object.

Writing and updating many rows efficiently

If you do this:

db = connect('mols.db')
for mol in molecules:
    db.write(mol, ...)

the database will make sure that each molecule is written to permanent storage (typically a harddisk) before it moves on to the next molecule. This can be quite slow. To speed this up, you can write all the molecules in a single transaction like this:

with connect('mols.db') as db:
    for mol in molecules:
        db.write(mol, ...)

When the for-loop is done, the database will commit (or roll back if there was an error) the transaction.

Similarly, if you want to update() many rows, you should do it in one transaction:

with db:
    for id in ...:
        db.update(id, foo='bar')

Writing rows in parallel

Say you want to run a series of jobs and store the calculations in one database:

for name in many_molecules:
    mol = read(name)
    calculate_something(mol)
    db.write(mol, name=name)

With four extra lines (see the reserve() method):

for name in many_molecules:
    id = db.reserve(name=name)
    if id is None:
        continue
    mol = read(name)
    calculate_something(mol)
    db.write(mol, id=id, name=name)

you will be able to run several jobs in parallel without worrying about two jobs trying to do the same calculation. The reserve() method will write an empty row with the name key and return the ID of that row. Other jobs trying to make the same reservation will fail. While the jobs are running, you can keep an eye on the ongoing (reserved) calculations by identifying empty rows:

$ ase db many_results.db natoms=0

More details

Use this function for getting a connection to a database:

ase.db.core.connect(name, type='extract_from_name', create_indices=True, use_lock_file=True, append=True, serial=False)

Create connection to database.

name: str

Filename or address of database.

type: str

One of ‘json’, ‘db’, ‘postgresql’, (JSON, SQLite, PostgreSQL). Default is ‘extract_from_name’, which will guess the type from the name.

use_lock_file: bool

You can turn this off if you know what you are doing …

append: bool

Use append=False to start a new database.

Here is a description of the database object:

class ase.db.core.Database(filename=None, create_indices=True, use_lock_file=False, serial=False)

Base class for all databases.

Database object.

serial: bool

Let someone else handle parallelization. Default behavior is to interact with the database on the master only and then distribute results to all slaves.

write(atoms, id=None, key_value_pairs={}, data={}, **kwargs)

Write atoms to database with key-value pairs.

atoms: Atoms object

Write atomic numbers, positions, unit cell and boundary conditions. If a calculator is attached, write also already calculated properties such as the energy and forces.

key_value_pairs: dict

Dictionary of key-value pairs. Values must be strings or numbers.

data: dict

Extra stuff (not for searching).

id: int

Overwrite existing row.

Key-value pairs can also be set using keyword arguments:

connection.write(atoms, name='ABC', frequency=42.0)

Returns integer id of the new row.

reserve(**key_value_pairs)

Write empty row if not already present.

Usage:

id = conn.reserve(key1=value1, key2=value2, ...)

Write an empty row with the given key-value pairs and return the integer id. If such a row already exists, don’t write anything and return None.

update(id, atoms=None, delete_keys=[], data=None, **add_key_value_pairs)

Update and/or delete key-value pairs of row(s).

id: int

ID of row to update.

atoms: Atoms object

Optionally update the Atoms data (positions, cell, …).

data: dict

Data dict to be added to the existing data.

delete_keys: list of str

Keys to remove.

Use keyword arguments to add new key-value pairs.

Returns number of key-value pairs added and removed.

metadata

Dictionary

get_atoms(selection=None, add_additional_information=False, **kwargs)

Get Atoms object.

selection: int, str or list

See the select() method.

add_additional_information: bool

Put key-value pairs and data into Atoms.info dictionary.

In addition, one can use keyword arguments to select specific key-value pairs.

get(selection=None, **kwargs)

Select a single row and return it as a dictionary.

selection: int, str or list

See the select() method.

select(selection=None, filter=None, explain=False, verbosity=1, limit=None, offset=0, sort=None, include_data=True, columns='all', **kwargs)

Select rows.

Return AtomsRow iterator with results. Selection is done using key-value pairs and the special keys:

formula, age, user, calculator, natoms, energy, magmom and/or charge.

selection: int, str or list

Can be:

• an integer id

• a string like ‘key=value’, where ‘=’ can also be one of ‘<=’, ‘<’, ‘>’, ‘>=’ or ‘!=’.

• a string like ‘key’

• comma separated strings like ‘key1<value1,key2=value2,key’

• list of strings or tuples: [(‘charge’, ‘=’, 1)].

filter: function

A function that takes as input a row and returns True or False.

explain: bool

Explain query plan.

verbosity: int

Possible values: 0, 1 or 2.

limit: int or None

Limit selection.

offset: int

Offset into selected rows.

sort: str

Sort rows after key. Prepend with minus sign for a decending sort.

include_data: bool

Use include_data=False to skip reading data from rows.

columns: ‘all’ or list of str

Specify which columns from the SQL table to include. For example, if only the row id and the energy is needed, queries can be speeded up by setting columns=[‘id’, ‘energy’].

count(selection=None, **kwargs)

Count rows.

See the select() method for the selection syntax. Use db.count() or len(db) to count all rows.

delete(ids)

Delete rows.

Metadata

It’s can be useful to add a discription of the data that’s in a database. This description will be used when browsing the database with a web-browser. You can add the desciption using the metadata attribute:

>>> db.metadata = {
...     'title': 'Project 1',
...     'key_descriptions':
...         {'v0': ('Voltage', 'Longer description ...', 'V')},
...     'default_columns': ['id', 'formula', 'v0']}

ASE already knows all about the following keys:

key	short description	long description	unit
id	ID	Uniqe row ID
age	Age	Time since creation
formula	Formula	Chemical formula
pbc	PBC	Periodic boundary conditions
user	Username	Username
calculator	Calculator	ASE-calculator name
energy	Energy	Total energy	eV
natoms	Number of atoms	Number of atoms
fmax	Maximum force	Maximum force	eV/Å
smax	Maximum stress	Maximum stress on unit cell	eV/ų
charge	Charge	Net charge in unit cell	|e|
mass	Mass	Sum of atomic masses in unit cell	au
magmom	Magnetic moment	Magnetic moment	μ_B
unique_id	Unique ID	Random (unique) ID
volume	Volume	Volume of unit cell	ų

You can also write/read to/from JSON using:

$ ase db proj1.db --set-metadata metadata.json
$ ase db proj1.db --show-metadata > metadata.json

External Tables

If the number of key_value_pairs becomes large, for example when saving a large number of features for your machine learning model, ASE DB offers an alternative way of storing them. Internally ASE can create a dedicated table to store groups of key_value_pairs. You can store a group of key_value_pairs in a separate table named features by:

>>> atoms = Atoms()
>>> no_features = 5000
>>> feature_dict = dict(('feature' + str(i), i) for i in range(no_features))
>>> id = db.write(atoms, external_tables={'features': feature_dict})

Values stored in external tables can be accessed using:

>>> row = db.get(id=id)
>>> f1 = row['features']['feature1']
>>> f4999 = row['features']['feature4999']

Running a PostgreSQL server

With your PostgreSQL server up and running, you should run the following command as the postgres user:

$ sudo -u postgres psql

Then you create an ‘ase’ user and one database for each project you have:

postgres=# create user ase login password 'pw';
postgres=# create database project1;
postgres=# create database project2;

Show databases and quit:

postgres=# \l
postgres=# \q

You should now be able to query the database using a URI like postgresql://user:pw@host:port/dbname:

$ PROJ1=postgresql://ase:pw@localhost:5432/project1
$ ase db $PROJ1

If you have some data in a data.db SQLite3 file, then you can insert that into the PostgreSQL database like this:

$ ase db data.db --insert-into $PROJ1

Now you can start the Flask-app ase.db.app.app with any WSGI compatible server.

Note

Please review the code carefully before exposing the ase.db.app to the internet or bad things could happen.

Running a MySQL server

ASE DB can also be run with a MySQL server. First, we need to get the MySQL server up and running. There are many online resources describing how to to that, but on a Ubuntu system the following should work:

$ sudo apt-get install mysql-server
$ sudo mysql_secure_installation

Then we need to check if the server is running:

$ systemctl status mysql.service

if it is not running you can start the service by:

$ systemctl start mysql.service

Note that on some Linux distributions mysql.service should be replaced by mysqld.service.

Once the service is running, we can enter the MySQL shell:

$ mysql -u root -p

where we assume that there is a user named root, that will be prompted for a password. Now, we can create a user:

mysql> CREATE USER 'ase'@'localhost' IDENTIFIED BY 'strongPassword';

and then a database for our project:

mysql> CREATE DATABASE my_awesome_project;

We need to give the ase user privileges to edit this database:

mysql> GRANT ALL PRIVILEGES ON my_awesome_project.* TO 'ase'@'localhost' IDENTIFIED BY 'strongPassword';

From a Python script we can now connect to the database via

>>> mysql_url = 'mysql://ase:strongPassword@localhost:3306/my_awesome_project'
>>> connect(mysql_url)