Download presentation
Presentation is loading. Please wait.
Published byJulian Woods Modified over 9 years ago
1
Genetic algorithm - Monte Carlo hybrid method for finding stable geometries of atomic clusters Application to carbon clusters Nazım Dugan, Şakir Erkoç METU, Physics Department
2
Outline Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion
3
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion
4
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion GA Overview Atomic systems >> 3N degrees of freedom (with empirical potentials) Minimum potential energy >> maximum stability Find global minimum on the Potential energy hypersurface Fitness criteria >> Total potential energy of the system Encoding : 111010001010100101000101 CTGA....
5
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion Genetic Operations Mutation Natural selection Reproduction Crossover 100110010010100110101100........ 101110010010100100101100........ 0011001001 00011001000 10011001001 11010001001 10011001101
6
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion -33 eV -28 eV-32 eV-31 eV-26 eV Genetic Operations Mutation Natural selection Reproduction Crossover 111010001010100101000101
7
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion Genetic Operations Mutation Natural selection Reproduction Crossover
8
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion 100110010010100110101100 001110100011010000101001 100110010010010000101001 Genetic Operations Mutation Natural selection Reproduction Crossover
9
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion We have to decide on Population size too low >> not enough diversity !!! too high >> not necessary !!! Mutation rate too low >> lose diversity !!! too high >> mutants !!! Reproduction type asexual sexual >> how to combine individuals Natural selection type not big deal (select better with higher probability) elitism
10
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion
11
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion random walk of atoms if (ΔE < 0) accept if (ΔE > 0) check convergence if rand(0-1) < exp(- ΔE / kT) accept LOOP
12
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion
13
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion search over all phase space >> local optimization between GA steps >> search over local minima geometric mutation operations Rotation Atom permutation Geometric crossover (Deaven - Ho)
14
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion
15
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion Generate N clusters randomly MC local optimization Natural selection -29.6342 eV-28.9163 eV-33.3887 eV-28.8877
16
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion Reproduction Mutation (Rotation) Mutation (Shrink) MC local optimization (1st step)
17
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion C 20 animation
18
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion Parallelization Distribute individuals to available nodes MC local optimization takes much more time >> all nodes do MC GA operations takes no time >> master node gathers individuals and do GA ~100 % efficiency !!!
19
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion
20
Genetic Algorithms (GAs) Classical Monte Carlo (CMC) Application of GAs to atomic clusters Method overview Results and discussion n = 11n = 16 n = 20 shrink 0.5 n = 38 Shrink 0.85 Carbon clusters (Tersoff - Brenner potential energy function) n = 12n = 14 n = 22 shrink 0.5 n = 32 Shrink 0.8 n = 19
21
References R.L. Johnston, Dalton Trans., 4193(2003) B. Hartke, Chem. Phys. Lett. 240, 560(1995) S.K. Gregurik, M.K. Alexander, B. Hartke, J. Chem Phys. 104, 2684(1996) S. Hobday, R. Smith, J. Chem. Soc., Faraday Trans. 93, 3919(1997) M. Iwamatsu, J. Chem Phys. 112, 10976(2000) J. Zhao, R. Xie, J. Comp. Theoretical Nanoscience 1, 117(2004) D.M. Deaven, K.M. Ho, Phys. Rev. Lett. 75, 288(1995) Thank you
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.