Slide 1
Slide 1 text
Yonsei University Graduate Class
Energy Materials: Design, Discovery and Data
Exploring the Materials Hyperspace
Dan Davies
PhD Student
University of Bath
https://wmd-group.github.io @danwdavies
Slide 2
Slide 2 text
About Me
• MChem University of Bath
• PhD student at the CSCT
• Interested in high-throughput
computational screening and design of
materials
Sustainable
Centre for
Chemical Technologies
Slide 3
Slide 3 text
About Me
Slide 4
Slide 4 text
I
Talk Outline: The Materials Hyperspace
1. Data for Materials Discovery
2. Exploring Chemical Space
3. Interacting with Databases
II
III
Slide 5
Slide 5 text
Question
How were the vast majority
of materials we use today
discovered?
Slide 6
Slide 6 text
Traditional Materials Discovery
• Trial and error
• Trial and improvement
“I have not failed, I have just
found 10,000 ways that
won’t work.”
Slide 7
Slide 7 text
Traditional Materials Discovery
Teflon®
Silly putty
Safety glass
Slide 8
Slide 8 text
Traditional Materials Discovery
Teflon®
“A white solid material
was obtained, which was
supposed to be a
polymerized product.”
www.whitetrout.net/Chuck/Teflon/teflon.htm
Slide 9
Slide 9 text
Traditional Materials Discovery
Teflon® “It is insoluble in cold and hot
water, acetone, Freon 113,
ether, petroleum ether,
alcohol, pyridine, toluene ethyl
acetate, concentrated sulfuric
acid, glacial acetic acid,
nitrobenzene, isoanyl alcohol,
ortho-dichlorobenzene, sodium
hydroxide, and concentrated
nitric acid.”
www.whitetrout.net/Chuck/Teflon/teflon.htm
Slide 10
Slide 10 text
Getting Materials to Market
Time between invention and widespread
commercialisation of materials:
Slide 11
Slide 11 text
Modern Materials Discovery
A. Agrawal & A. Choudhary, APL Materials, 2016, 4
• Data from experiment and calculation is now
being generated at an incredibly fast rate
• This has allowed
for the
emergence of
“Big Data”
driven science
Slide 12
Slide 12 text
The 4 Vs of Big Data
K. Rajan, Annu. Rev. Mater. Res., 2015, 153-169
Slide 13
Slide 13 text
The ICSD – Experimental Data
Inorganic Crystal Structure
Database
• Large database
crystallographic data
• ~187,000 crystal
structures
Slide 14
Slide 14 text
The ICSD – Experimental Data
Slide 15
Slide 15 text
Experimental vs. Computed Properties
“Real”/measured
properties of materials
No great databases
useful for data-driven
approach L
(with exception of
crystallographic data)
Simulated/predicted
properties of
materials
Some good emerging
databases to choose
from J
Slide 16
Slide 16 text
Data From Calculations
Slide 17
Slide 17 text
The NoMaD Repository
https://nomad-coe.eu/
Novel Materials Discovery
• Contains input and output
files from electronic
structure calculations
• 3,300,000 entries
• Anyone can upload à
Inhomogeneous data
Slide 18
Slide 18 text
The Materials Project
• Uses ICSD as primary
input source
• 67,000 entries
• All calculations similar
à homogeneous data
https://materialsproject.org/
Slide 19
Slide 19 text
Using the Available Data
Calculated From Database
Slide 20
Slide 20 text
Using the Available Data
Machine learning
A type of artificial intelligence that provides computers
with the ability to learn without being explicitly
programmed.
Slide 21
Slide 21 text
Using the Available Data
Machine learning – Total Energies
Using element
properties such as
atomic mass, #
valence e-, ionisation
potential etc. along
with connectivity
within material
Machine learning algorithm from https://arxiv.org/pdf/1608.04782.pdf
Slide 22
Slide 22 text
Using the Available Data
Machine learning – Band gaps
Using only
element
composition as
input – no other
information
Slide 23
Slide 23 text
Talk Outline: The Materials Hyperspace
1. Data for Materials Discovery
2. Exploring Chemical Space
3. Interacting with Databases
Slide 24
Slide 24 text
Question
How many unique
materials* are known
today?
*Inorganic compounds
Slide 25
Slide 25 text
Counting Known Compounds
~187,000
Hard to quantify – no definitive list exists
~3,300,000
~67,000
Lots of duplicates
Many hypothetical
Uses ICSD as input
But not exclusively
Not finished
All ‘real’ (mostly)
Some duplicates
Slide 26
Slide 26 text
Counting Possible Compounds
http://hackingmaterials.com/2013/11/14/the-scale-of-materials-design/
• You have a list of 50 chemical elements and a
10x10x10 grid
• You can put 30 atoms of any element anywhere
on the grid to make a unit cell
H He Li Be B C N O F Ne
Na Mg Al Si P S Cl Ar K Ca
Sc Ti V Cr Mn Fe Co Ni Cu Zn
Ga Ge As Se Br Kr Rb Sr Y Zr
Nb No Tc Ru Rh Pd Ag Cd In Sn
I
Slide 27
Slide 27 text
Counting Possible Compounds
print(’the value of x is {0}’.format(x))
Format statements:
Q. What is the purpose of using the second type of format
statement? (Try replacing a {:.2E} with a simple {0} in your
notebook to see.)
print(’the value of x is {:.2E}’.format(x))
I
Slide 28
Slide 28 text
Mapping Out Chemical Space
• Clearly this approach is a
non-starter; we need to
simplify things
• One way is to combine
elements in their known
oxidation states to make
binary, ternary, quaternary
combinations…
Slide 29
Slide 29 text
Mapping Out Chemical Space
SMACT
Semiconducting Materials by Analogy and Chemical Theory
• Combine elements together in
their known oxidation states
exhaustively
• Only allows certain
combinations based on certain
rules e.g. charge neutrality
Sn2+ O2- I-
True
Ratios = [(2,1,2),
(3,2,2)]
Charge neutral
combinations
possible?
(stoichiometry
threshold = 3)
Slide 30
Slide 30 text
Mapping Out Chemical Space
SMACT
Semiconducting Materials by Analogy and Chemical Theory
• Combine elements together in
their known oxidation states
exhaustively
• Only allows certain
combinations based on certain
rules e.g. charge neutrality
Mn7+ F- I-
False
Ratios = []
Charge neutral
combinations
possible?
(stoichiometry
threshold = 3)
II
Slide 31
Slide 31 text
Mapping Out Chemical Space
all_elements = smact.element_dictionary()
Setting up a dictionary
Dictionary of
element objects
Built in smact function
Q. Can you spot the other
smact function that
appears in the same
cell?
II
Slide 32
Slide 32 text
Mapping Out Chemical Space
for i, ele_a in enumerate(elements):
Using the enumerate function
is the same as
for i in range(len(elements)):
ele_a = elements[i]
II
Slide 33
Slide 33 text
Mapping Out Chemical Space
for j, ele_b in enumerate(elements[i+1:]):
List slicing
[where to start]:[where to stop]
Q. What does putting
this within our inner
loop achieve?
II
Slide 34
Slide 34 text
Talk Outline: The Materials Hyperspace
1. Data for Materials Discovery
2. Exploring Chemical Space
3. Interacting with Databases
Slide 35
Slide 35 text
Question
How can we access
materials data?
Slide 36
Slide 36 text
Accessing Materials Data
• Web Browser e.g. Materials Project
• Data dump e.g.
Computational Materials Repository
• Restful API e.g. Materials Project MAPI
Slide 37
Slide 37 text
RESTful API
• Representational State Transfer Application
Programming Interface
• Built around resources and how they are
accessed
• Resources == objects in object oriented
programming!
Slide 38
Slide 38 text
Materials API
• Register:
MaterialsProject.org
• Copy and paste your API key:
• m = MPRester(" ")
III
Slide 39
Slide 39 text
Materials API
General usage
III
data = m.query(criteria, properties)
API query function Criteria of the
entries we’re
interested in
Properties we want
to get back
Returns a list of dictionaries
[ {property_1 : value, property_2: value},
{property_1 : value, property_2: value} ]
Entry 1
Entry 2
Slide 40
Slide 40 text
Materials API
Getting clever with criteria
III
criteria = { ’nelements’: 2,
‘elements’:
{‘$in’: [‘Co’, ‘Fe’]} }
Simple Python
dictionary usage
MongoDB operator
usage (as python strings)
Also accepts $gt, $lt, $eq, $all, $nin,
$exists etc.
Slide 41
Slide 41 text
Summary – Key Points
• Materials data is being generated at very fast
rate
• Emerging efforts to store and organise the
data can speed up materials discovery
• Python + modern databases = huge amounts
of materials data at your fingertips
• There are vast areas of the chemical
landscape that remain totally unexplored