stut pedon arm: in si de si f mel
TRANSCRIPT
![Page 1: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/1.jpg)
St u t P ed on A r m : In Si Desi f M e l
José A. Flo -LivUni s of B
05.09.2018, Halle (Saale), Germany.
![Page 2: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/2.jpg)
Is it possible to predict the crystal
structure of a substance ?
2
“One of the continuing scandals in the physical sciences is that it remains impossible to predict the structure of even the simplest crystalline solids from a knowledge of their composition.” John Maddox, Nature 1988
![Page 3: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/3.jpg)
3
Carbonpolymorphs
Some substances have more than one crystal structure !
![Page 4: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/4.jpg)
Polymorphism is important in pharmaceutical industry, drugs, organic and inorganic compounds, viruses, of course DNA.
4
Materials Science Other Sciences
Arrangement of atoms in
Ritonavir (HIV/AIDS) case: form I during development but once in market form II was found more stable. Loses of millions US dollars.
Polymorphism
![Page 5: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/5.jpg)
5
Why we want to know the structure or predict it?
The arrangement of atoms is the most important piece of information about a material.
Fundamental tool widely use
➔ To interpret experimental data of poor quality.➔ To study materials under thermodynamic conditions beyond experimental limits.➔ As a driving force for materials design: artificial intelligence, neural networks and machine learning.
From the structure one can derive models as well, obtain observables (physical properties) of molecules, crystals, etc, using quantum-mechanical methods.
![Page 6: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/6.jpg)
6
Material's properties are linked to atomic structure
Structure related properties
![Page 7: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/7.jpg)
Table of Contents:
I) In od on n e t
II) Al o t sRandom searchThermodynamic methodsGenetic basedThe minima hopping method
III) Bey s uc p e c o : Ap ic o sComputational Materials ScienceExotic ChemistryFingerprints and machine learning
IV) Sum y
![Page 8: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/8.jpg)
8
➔ Structure prediction ↔ global optimization of Gibbs free energy.
➔ It is a 3N -dim potential energy or 3N + 9-dim enthalpy surface
➔ Stable configurations are minima on the potential energy surface:
➔ Structures are minima on a surface (enthalpy) with real
The global minimum: Minimizing a target function
![Page 9: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/9.jpg)
9
What is the Potential Energy Surface? (PES)
A potential energy surface is a mathematical function that gives the energy of a system as a function of its geometry.● Energy landscape depends on chemical space● Accessible for molecular and crystals systems● Energy evaluated using force-fields, tight-binding, DFT and ML
Energy landscapes
![Page 10: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/10.jpg)
10
Potential energy surface
High energy structure
Global minimum
Saddle point
Local minima
Saddle point
Energy
Reaction coordinate
Reaction coordinate
![Page 11: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/11.jpg)
11
Potential energy surfaceHigh energy
structure
Global minimumSaddle
“structure”
Local minima
![Page 12: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/12.jpg)
Table of Contents:
I) In od on n e t
II) Al o t s
> Random searchThermodynamic methodsGenetic basedThe minima hopping method
III) Bey s uc p e c o : Ap ic o sComputational Materials ScienceExotic ChemistryFingerprints and machine learning
IV) Sum y
![Page 13: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/13.jpg)
13
➔ It is simple, easy to implement, surprisingly successful: robust.
But, how random is random?
➔ Implement penalty radii➔ Insight from chemical composition➔ Constrain or exploit symmetry➔ Despite all this, it is a ”Shake and bake” method and is not suitable for many atoms.
Random Search
![Page 14: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/14.jpg)
14
➔ Width of catchment basins correlates well with well(s) energy.➔ Successful implementation: Ab initio Random Structure Searching (AIRSS).*
* Pickard and Needs, PSI-k Highlight 100, (2010).
Random search: AIRSS
![Page 15: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/15.jpg)
Table of Contents:
I) In od on n e t
II) Al o t sRandom search
> Thermodynamic methodsGenetic basedThe minima hopping method
III) Bey s uc p e c o : Ap ic o sComputational Materials ScienceExotic ChemistryFingerprints and machine learning
IV) Sum y
![Page 16: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/16.jpg)
16
● Method to generate a Boltzmann distribution. ● Probability of finding configuration X is proportional to● At sufficiently low T, the ground state will be the dominant configuration.● Works well for single-funnel energy landscapes.
➔ Mimics the process to reduce defects in solids: higher barriers can be crossed at high temperature.
➔ System is propagated by molecular dynamics at high T.
➔ Controlled cooling: crystallization to the ground state according to the Boltzmann distribution.
➔ Guaranteed to find the global minimum if thermodynamic equilibrium is reached.
Thermodynamic methods
Simulated annealing
![Page 17: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/17.jpg)
17
➔ In practice rather slow, depending on the annealing schedule.
➔ Often gets stuck in a funnel or local minimum.
Simulated annealing
![Page 18: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/18.jpg)
18
➔ Metadynamics is performed in the space defined by collective variables S.➔ The dynamics is driven by the free energy and is biased by a history-dependent potential
constructed as a sum of Gaussians centered along the trajectory of S.➔ Not useful for crystal discover. ➔ But this approach can compute the free energy, exploring reaction pathways and
accelerating rare events.
Simulated annealing
![Page 19: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/19.jpg)
Table of Contents:
I) In od on n e t
II) Al o t sRandom searchThermodynamic methods
> Genetic basedThe minima hopping method
III) Bey s uc p e c o : Ap ic o sComputational Materials ScienceExotic ChemistryFingerprints and machine learning
IV) Sum y
![Page 20: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/20.jpg)
20
● Mimic a Darwinistic evolution.
● Population of individuals (solutions Xi) with genes, represented by binary strings.
● Genetic operations: Gene crossing, mutation, selection, mating, recombine, etc.
● Survival of the fittest, bad solutions are eliminated: Peppa pig !
● Iterate and keep population healthy: low energy and high gene diversity.
Genetic evolutionary based algorithms
![Page 21: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/21.jpg)
Genetic algorithms: “Qu n Dar s a D s e -Pix t”
Painful process
of generation:
E
![Page 22: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/22.jpg)
22
➔ Unnatural to map continuous atomic coordinates to binary strings.➔ Modern GAs operate directly on atomic coordinates.➔ Various methods to perform crossover in structures.➔ GA implementations: USPEX, XtalOpt, GASP, Abinit, etc.➔ GA and momenta: Particle swarm optimization (CALYPSO, firefly).
Genetic evolutionary based algorithms
![Page 23: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/23.jpg)
Table of Contents:
I) In od on n e t
II) Al o t sRandom searchThermodynamic methodsGenetic based
> Minima hopping method
III) Bey s uc p e c o : Ap ic o sComputational Materials ScienceExotic ChemistryFingerprints and machine learning
IV) Sum y
![Page 24: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/24.jpg)
24
Approach of the method
Use walker to explore the (3N − 3) + 6−dim enthalpy surface by visiting local minima efficiently, hopping from one minimum to the next.*
Strategy of the walker
➔ Perform hops that lead to low energy structures.
➔ Avoid revisiting local minima.
➔ Allow crossing already explored regions on PES.
➔ How to hop: random displacements, single ended saddle point, etc.
➔ Constant energy Molecular Dynamics is more efficient!
Minima Hopping Method (MHM)
* MHM: Goedecker Journal of Chemical Physics, 120, 9911, 2004
![Page 25: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/25.jpg)
● Reactant and product are neighboring minima and
chemical reactions are transitions of barriers
connecting them.
● Energy conservation limits surmountable barrier
height by Ekin.
● Low Ekin in MD for hops into low neighboring minima !
-- The difference in activation energy between two reactions of the same family is proportional to the difference of their enthalpy of reaction --
“Exothermic reactions have a low activation energy”
Optimal moves
![Page 26: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/26.jpg)
26
Outer loop
➔ Accepting or rejecting.➔ Preference for lower energy, Ediff➔ Dynamic Ediff 50% accepted. Eventually
accept high energy structures.➔ Feedback on Ekin based on history
Inner loop
● Perform MD escape trials followed by local geometry relaxation.
● Ekin continuously adjusted
The algorithm
![Page 27: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/27.jpg)
27
Minima hopping for a 512 atom NaCl cluster
❖ Global minimum: 8 by 8 by 8 cube❖ Wrong funnel: 7 by 8 by 9 (=504) cuboid with defects
Example of the feedback mechanism
![Page 28: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/28.jpg)
28
Curvatures of eigenmodes of local minima correlate with corresponding energy barrier heights. Align initial MD velocity vectors along soft modes!
● Efficiency depends highly on the moves.● Optimized initial velocity of MD trajectory is crucial.● Soft modes lead to low barriers -Softening-
Optimizing the Moves
![Page 29: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/29.jpg)
Table of Contents:
I) In od on n e t
II) Al o t sRandom searchThermodynamic methodsGenetic basedThe minima hopping method
III) Bey s uc p e c o : Ap ic o sComputational Materials ScienceExotic ChemistryFingerprints and machine learning
IV) Sum y
![Page 30: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/30.jpg)
30
Find element that “do mix” (miscible)● Binaries:● Ternaries:● Quaternaries:
● Find possible stoichiometries (convex hull).● Find correct crystal structure.● Non-ambient conditions.
(pressure/temperature/strain)
Se r s a p ti l po b !
To accelerate materials discovery: ➔ High-Throughput, database(s), etc. ➔ Select few prototype and substitute sites. ➔ Constrain the space search.
Materials discovery: The Chemical Space
![Page 31: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/31.jpg)
31
Materials database(s)
So n il q i at e of d a s ...
![Page 32: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/32.jpg)
32
Can we do, not say better but differently ? 1) Fully ab-initio, chemically-unconstrained search for targeted properties.
2; Combine: machine learning structure prediction
3; Stick to the basics; ask good experimentalists for a very specific problem.
![Page 33: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/33.jpg)
Table of Contents:
I) In od on n e t
II) Al o t sRandom searchThermodynamic methodsGenetic basedThe minima hopping method
I ) Bey s uc p e c o : Ap ic o sComputational Materials Science
> Exotic ChemistryFingerprints and machine learning
IV) Sum y
![Page 34: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/34.jpg)
34
Fingerprints, machine learning and superconductors
![Page 35: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/35.jpg)
Exo c is y
![Page 36: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/36.jpg)
36
260 Kel (-13.5 0C) su c uc
* Hemley, et al. ArXiv, (23 August 2018), posted: “Evidence for superconductivity above 260 K in lanthanum superhydride at megabar pressures”
The best structural candidate to explain the Tc (clathrate type)
formed by LaH6??
![Page 37: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/37.jpg)
Summary
CSP methodsCrystal structure prediction is a fundamental tool in diverse fields, from bio, pharma, to materials.
Random searchEasy to implement and to use, good starting point for many problems.
Genetic basedGNU packages available, relatively easy to start with, requires more expertise.
Therm. methodsMost complicated methods, not so easy to use, requiere great expertise to be useful, but are powerful.
MHMEasy to implement, also GNU distributed with BigDFT. Powerful and require little knowledge.
Beyond CSP CSP + high throughput, CSP + machine learning, AI, neural networks potentials, reaction pathways, nucleation.
37
![Page 38: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/38.jpg)
Ac ow g e t d di
FIRST
SECOND
LAST
Friends and collaborators: Antonio, Max, Migle, Kay, Sangeeta, Deb, Daniele, Lilia, Luigi, Gianni, Prof.: Marques, Gross, and Goedecker.
My computing grants: (S752) in Piz Daint and (pn69yi) in SuperMUC
38
![Page 39: Stut Pedon Arm: In Si De si f Mel](https://reader036.vdocuments.net/reader036/viewer/2022081403/6292355637b9ee3d7b7d1c5f/html5/thumbnails/39.jpg)
T Y !
www.fysik-aztek.net
Un piacere!
Any questions?
You can find more about our research at
Do you want a copy of the slides?
<< Snap a picture of this slide and you have it !QR code