[an error occurred while processing this directive]
Basics of constructing cluster expansions for alloy modeling
Gus Hart
Physics
Northern Arizona University
Phone: (928)-523-0426
Email: Gus.Hart ---> nau.edu
Lectures
- Cluster Expansions: Treating the effects of strain:
pdf (2.7MB) | ppt (4.3MB) | ppt (4.3MB, printer-friendly, reduced number of animations) | key.zip (4MB, the original Keynote presentation, zipped)
- Building Model Hamiltonians via an Evolutionary Approach:
pdf (1623 K) | ppt (3686 K) | ppt (3392 K)
| key.zip (3.4 MB, the original Keynote presentation, zipped)
Computational Laboratory
Details about Lectures and Computer Lab
Lecture abstract: Basic formalism behind alloy cluster expansions, that is the generalized Ising model, will be discussed. We will learn how to calculate "input data" for fitting to the model and then use the resulting model Hamiltonian, a cluster expansion, to make ground state predictions, to do thermodynamic Monte Carlo simulations.
Lab activities:
(i) importance of strain in phase separating systems
(ii) how to parameterize strain for use with a cluster expansion
(iii) Thermodynamic MC for phase transitions
(iv) Constructing CEs using an evolutionary approach
Cluster Expansions: Treating the effects of strain
- Generalized Ising model
- An Illustration: Cluster expansion (Ising)
- Cluster Expansion
- Coherent semi-infinite slabs
- Precipitates!
- Coherent semi-infinite slabs
- The resulting strain energy is "long-range"
- What do we mean by that?
- How is it that the Ising model (cluster expansion) can’t handle this?
- How can this be fixed?
- Superlattices...
- Consider a 2n-layer superlattice AnBn
- What happens as n gets large? (Or as k gets small if we think in k-space?)
- Non-analyticity
- pdf (2.7MB) | ppt (4.3MB) | ppt (4.3MB, printer-friendly, reduced number of animations) | key.zip (4MB, the original Keynote presentation, zipped)
Building Model Hamiltonians via an Evolutionary Approach
- The argument for "fancy"
- Goal: A fast, but quantitatively accurate, Hamiltonian
- Such models are complex, almost by definition generally not even the "shape"of the model is simple
- That is, the value of parameters is not the issue, which parameters to use in the model (or even how many!) is essentially unguessable
- Fancy or not, we need an approach that "works"
- An Illustration: Cluster expansion (Ising)
- How do we truncate? Determine the model's "shape?"
- Selecting the multi-body interaction types (MBITs)
- Optimization (search) problem:
- (Actually, the problem is even worse for surfaces...)
- How to pick the terms
- By physics "intuition" dangerous (and physics shouldn't be an "art")
- Formal hierarchy (converges way too slow to be practical)
- Simulated annealing (didn't work highly correlated problem)
- Systematic "guessing"hierarchal approach
- Something "fancier"(anything that works!)
- Formal hierarchy
- Evolutionary optimization
- Good approach for some highly correlated problems
- Not "too fancy"if it's the simplest thing that works
- Make a cool figure: on the left, b/w vs. color genomes
- Fade in on right 0010011111001111
- But does it work?
- Still have problems with local minima...
- "Lock-out"makes the approach completely robust
- Do we really need all this accuracy in the fitting/construction? Does it really make a difference?!
- Now for the fun part... (Today's lab)
- A simple view of the problem
- The specific problem for today
- Computing the energy
-
- pdf (1623 K) | ppt (3686 K) | ppt (3392 K) | key.zip (3.4 MB, the original Keynote presentation, zipped)
Computational Laboratory
Download the collection of tutorial materials:
Or the individual files: