The Nuts and Bolts of First-Principles Simulation Durham, 6th-13th December : Is it running properly? CASTEP Developers’ Group with support from the ESF  k Network

Nuts and Bolts : Is it running properly? 2 Outline  Why you have to read the output file  Sample output  The road to hell  Look after the pennies  Completely wrong  Right and wrong  Graphic detail  Summary

Nuts and Bolts : Is it running properly? 3 What is the question?  We have to eliminate technical failures and separate artefacts from the proper predictions of DFT theory within an approximation (LDA, GGA)  Goes much further than just checking the basics, because testing is the bedrock of good simulation studies (see Lecture 10)  Here though we start with the basics

Nuts and Bolts : Is it running properly? 4 Scrutinise the output file!  But I’ve got Materials Studio… CASTEP sets all the defaults….  But I have already set up the job properly…  But it finished ok without crashing…  But the results look ok, what else is there..?

Nuts and Bolts : Is it running properly? 5 But I’ve got Materials Studio…  Defaults (interface or CASTEP) may not be appropriate  Choices made on your behalf: are they what you want?  The program tells you what it is doing and what’s going wrong

Nuts and Bolts : Is it running properly? 6 But I have already set up the job properly… All of the previous, and  That doesn’t mean it has run properly  You have to verify that your intentions were realised: you might have asked the impossible  Are the results physically reasonable?

Nuts and Bolts : Is it running properly? 7 But it finished ok without crashing… All of the previous, and  Finished because.. The task was finished and the tolerances met or it ran out of iterations?  Was it efficient? But the results look ok, what else is there..?

Nuts and Bolts : Is it running properly? 8 The output file (.out.cst ) Input and defaults Initialisation Electronic minimisation (then MD, geometry optimisation) Calculate properties (Eigenvalues, forces and stress, band structure…) Finish

Nuts and Bolts : Is it running properly? 9 Sample output: startup | | | It is no longer just Cambridge... | | it is no longer just serial... | | it is no longer just for total energies... | | | | but for marketing and legal reasons it is still | | | | CCC AA SSS TTTTT EEEEE PPPP | | C A A S T E P P | | C AAAA SS T EEE PPPP | | C A A S T E P | | CCC A A SSS T EEEEE P | | | | | | Welcome to Castep II.0, the DFT code for the | | professional user. | | | | Authors: | | M. Segall, P. Lindan, M. Probert, C. Pickard, | | P. Hasnip, S. Clark, M. Payne |

Nuts and Bolts : Is it running properly? 10 Startup | | | Copyright 2000,2001 | | | | Please cite | | | | A Real and Reciprocal Space Odyssey | | (Comp. Phys. Comms 2001) | | | | in all publications arising from your use of | | this fine product. | | |  Or perhaps M. C. Payne, M. P. Teter, D. C. Allan, T. A. Arias and J. D. Joannopoulos, Rev. Mod. Phys. 64 (1992) 1045  …but please acknowledge

Nuts and Bolts : Is it running properly? 11 Inputs defaults Warning: parameters_read - missing parameters file, using defaults Strategy found: 1 g-vector groups containing 1 nodes. Initialising basis set. Standard grid dimensions: Fine grid dimensions: Plane wave load balancing: max 1899 min 1863 average 1881 Warning: parameters_output - missing parameters file, using defaults ************************************ Title ************************ ***************************** General Parameters ****************** output verbosity : normal (1) write checkpoint data to : si_8atoms.check type of calculation : single point energy unlimited duration calculation *********************** Exchange-Correlation Parameters *********** using functional : Local Density Approximation (LDA) initialisation

Nuts and Bolts : Is it running properly? 12 ************************* Pseudopotential Parameters ************** pseudopotential representation : reciprocal space representation : reciprocal space **************************** Basis Set Parameters ***************** basis set accuracy : FINE finite basis set correction : automatic **************************** Electronic Parameters **************** number of electrons = 32 number of up spins = 16 number of down spins = 16 number of spin components = 1 number of bands = 19 ********************* Electronic Minimisation Parameters ********** Method: Treating system as metallic with ensemble DFT treatment of electrons, and number of SD steps = 1 and number of CG steps = 9 total energy convergence tolerance = E-04 Electron Volt eigen-energy convergence tolerance = E-05 Electron Volt periodically dump wavefunctions to : si_8atoms.wvfn *******************************************************************

Nuts and Bolts : Is it running properly? Unit Cell Real Lattice(Angstrom) Reciprocal Lattice(Angstrom-1) Lattice parameters(Angstrom) Cell Angles a = alpha = b = beta = c = gamma = Current cell volume = Angstrom^ Cell Contents Total number of ions in cell = 8 Total number of species in cell = 1 Max number of any one species = 8

Nuts and Bolts : Is it running properly? 14 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x Element Atom Fractional coordinates of atoms x x Number u v w x x x x Si x x Si x x Si x x Si x x Si x x Si x x Si x x Si x xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx No user defined ionic velocities Details of Species Mass of species in amu Si Files used for pseudopotentials: Si Si_00.recpot

Nuts and Bolts : Is it running properly? k-Points For BZ Sampling MP grid size for SCF calculation is Number of kpoints used = Number Fractional coordinates Weight Symmetry and Constraints There are no symmetry operations specified or generated for this cell Centre of mass is constrained Number of ionic constraints = 3 Set iprint > 1 for details of ionic constrains Number of cell constraints= 0 Cell constraints are: External pressure/stress (Gigapascal)

Nuts and Bolts : Is it running properly? <-- SCF SCF loop Energy Energy gain Timer <-- SCF Initial Final per atom (sec) <-- SCF <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF E E E <-- SCF <-- SCF Electronic minimisation  NB: “very-beta” test run and timings! Try this one yourself

Nuts and Bolts : Is it running properly? 17 The road to hell I thought I had Used the rPBE functional... Run it as a metal... Used the gamma point... Continued a run… Altered the lattice parameters… Used that other pseudopotential… Run without symmetry… Fed the dog my homework...  Or anything you care to mention. The truth is in the output file!

Nuts and Bolts : Is it running properly? 18 Look after the pennies If CASTEP ran what you wanted it to:  Are there any warning messages? Do you know if they are important or irrelevant?  Has it converged or just stopped?  Are all the obvious quantities reasonable? Total energy: negative? Expected value? Forces and stress: zero if they should be? Too large? Electronic energies: related to known values?

Nuts and Bolts : Is it running properly? 19 Completely wrong  Apart from the obvious: The job never started The job crashed  There are many ways a calculation can be rendered useless, but usually are avoided via systematic testing (Lecture 10) Inadequate plane-wave cutoff, k-point sampling Pseudopotential ghost states, poor transferability System size  However, within a single calculation things can still go wrong, a few examples follow

Nuts and Bolts : Is it running properly? 20 Right and wrong I  Although density mixing never does produce a smooth E vs. n curve (e.g. Cu unit cell)...

Nuts and Bolts : Is it running properly? 21 …right and wrong I  Sometimes it just doesn’t work (e.g isolated Carbon atom,  -point)

Nuts and Bolts : Is it running properly? 22 Right and wrong II  Symmetry Use of (point group) symmetry imposes constraints The supercell imposes the translational symmetry constraints Both can lead to the wrong answer

Nuts and Bolts : Is it running properly? 23 Right and wrong III  Load balancing On a parallel computer the node with the most work dictates overall performance Example (artificial!) 4 k-point calculation on three nodes, k-parallel Max. speedup is only 2X, not 3X

Nuts and Bolts : Is it running properly? 24 Graphic detail  Use graphics to see what you are calculating xo yo zo dx dy dz

Nuts and Bolts : Is it running properly? 25 Summary  The output file contains the basic information on the identity, health and sanity of your calculation  Here we have looked at a tiny fraction of the potential pitfalls  Always use/check the most basic data first  Plot graphs, use graphic images  If it can go wrong it will - but only when you are not looking!