Debyer¶
In short: Debyer can calculate diffraction pattern (from Debye’s scattering equation), pair distribution function and a few other statistics of an atomistic model.
The source code (C/C++) and installation instructions are on GitHub.
Debyer takes as an input a file with positions of all the atoms in the virtual sample (up to tens of millions of atoms, perhaps even more) and can output
- x-ray and neutron powder diffraction pattern, total scattering structure function,
- pair distribution/correlation function (PDF/PCF) and related functions (RDF, reduced PDF).
Diffraction patterns are calculated using the Debye scattering equation, hence the name.
A few use cases for Debyer can be found in the papers citing its website. Usually, the analyzed model
- is a result of molecular dynamics simulation,
- or is constructed using simple geometrical rules, with no interatomic potentials and no forces.
Debyer has two working modes: with and without periodic boundary conditions (PBCs). PBCs approximate a large (infinite) system. Otherwise, you have a finite atomic cluster or maybe a nano-grain in the (infinite) vacuum.
Debyer is not designed to calculate a diffraction pattern of a perfect crystal. It does not make sense to employ the Debye’s formula for a unit cell: you can easily get indexed pattern using other programs, such as PowderCell (not maintained since 1990s) or CCDC Mercury (any others?).
Note
If the math below doesn’t render properly, right-click on it and change
Math Renderer
in Math Settings
.
Direct-space patterns¶
Unlike in experiments, the pair distribution function (pair correlation function) is obtained directly from the atomic coordinates, not through the diffraction pattern. In other words, we do not emulate errors of experimental PDFs.
The definitions of PDF/RDF/PCF vary. Here, we stick to names and symbols from the Underneath the Bragg Peaks book (ISBN 0-08-042698-0):
- g(r) – atomic pair distribution function (PDF), a.k.a. pair correlation function (PCF), it converges to 1 as r increases.
- G(r) – reduced PDF (rPDF), which oscillates around 0,
- R(r) – radial distribution function (RDF), which goes up like parabola,
- ρ(r) – atomic pair density function, proportional to g(r), rarely used but listed here for completeness.
The average density \(\rho_0\) is either defined by the user
(--ro
, as #atoms / volume)
or determined automatically.
Average density is well-defined for a continuous system in the PBC (periodic boundary conditions), but not for a single grain in the vacuum. In the latter case G(r) will be distored. Although corrections for particular grain shapes have been proposed in the literature, they are not implemented in this program.
Weights b can be set as neutron scattering lengths, x-ray f(0) or equal 1 (see the section Scattering factors).
Debye scattering formula¶
Reciprocal space patterns (X-ray and neutron powder diffraction patterns) are calculated using the Debye scattering equation:
where
- \(Q\) is the scattering vector, called also momentum transfer vector (\(Q = \left|\boldsymbol{Q}\right| = 4 \pi \sin \theta / \lambda\) , where \(\theta\) is diffraction half-angle and \(\lambda\) is the wavelength),
- \(r_{ij}=\left|\boldsymbol{r}_i - \boldsymbol{r}_j\right|\) is the distance between atoms i and j,
- and \(f_i\) is the atomic scattering factor of i-th atom; in general it depends on \(Q\) and we should write it as \(f(Q)\) , but we don’t, to keep the notation simple.
Note
Different symbols are used in the literature. In particular, the \(q\) symbol is sometimes used instead of \(Q\) (with the same meaning). \(q\) may also be defined as \(q = 2\sin\theta / \lambda = Q / 2\pi\) . Or \(s\) may be used for the latter.
The best way to understand what this equation calculates, what assumptions and approximations are used, is to derive it.
Derivation¶
Derivation of the Debye formula can be found in many places. The standard citation is a German-language paper, P. Debye, Ann. Physik., 46 (1915) 809. Re-derivation can be found for example in Farrow & Billinge, Acta Cryst A65 (2009) 232 (pre-print).
Let’s re-derive it once more. We start with the amplitude of the scattered wave. If we ignore non-elastic scattering and assume that each photon is scattered only once, the amplitude is a simple sum over all atoms:
(i has two unrelated meanings here, index and the imaginary unit, hopefully it’s not too confusing).
The intensity of the scattered wave is given as
where the complex conjugate
Therefore, with the assumption that atomic form factors are real and \(f^{*}=f\) ,
The Debye scattering equation gives spherically averaged intensity. The averaging is similar to calculating the surface area of sphere. It is more elegant to use double integral, but knowing the circumference formula we can take a shortcut and use one integral:
Now, let \(\gamma\) be the angle between \(\boldsymbol{Q}\) and \(\boldsymbol{r}_{ij} \, (\equiv \boldsymbol{r}_i - \boldsymbol{r}_j)\) .
Avaraging the exponential part gives
which proves the Debye formula.
Modifications¶
Usually this formula is normalized by \(\frac{1}{N}\) or \(\frac{1}{N<f>^2}\) . TBC
include Debye-Waller factor? the correlated broadening factor for the atom pair (as mentioned in Farrow 2009)?
Cut-off¶
In this section, to simplify notation, we will consider monoatomic system. It can be easily generalized to multiple species.
The Debye formula allows us to directly calculate the intensity from an isolated particle. But in “infinite” bulk material (simulated in periodic boundary conditions) we must somehow limit the number of considered atomic pairs.
The simplest idea could be to pick a cut-off distance and limit the Debye formula to atomic pairs not further apart than this distance. But the termination effect would create a large sinusoid. So it is a common practice to add corrections for this effect.
Further calculations will be easier if we write the Debye formula as an integral,
where
Compensation¶
Let us compensate for the missing atomic pairs, approximating the structure beyond the cut-off distance \(r_c\) with a continuum:
The density of the continuum \(\rho\) is set to the average density of the structure, and
where N is the total number of atoms. We need to calculate the second addend.
Oops, it doesn’t converge.
It will converge if we subtract from I(Q) the intensity diffracted from a continuus system with density \(\rho\).
Since continuum does not add to diffraction (at non-zero Q), it should not harm to subtract \(I_{cont}\) from the right side of the approximation above. (I haven’t seen it explained like this in the literature, but this operation is quite common).
So now the correction is \(I_{cont}^{r>r_c} - I_{cont} = - I_{cont}^{r<r_c}\) ,
Finally,
That’s the simplest correction. It can be also applied without the analytical form above. If the summation is using eq. (1) (with histogram approximation, i.e. n(r) is counted in finite intervals), the alternative is to subtract ncont(r) from n(r) in each interval.
TODO: introduce structure factor S(Q). Would using S(Q) instead of I(Q) make things simpler?
Damping¶
The correction above works well enough for polycrystalline systems, but may not work for a single crystal. It should work fine if the pair correlation function is flat at the cut-off distance. If it is not flat, it is necessary to smooth it using a damping function.
In a few papers the sinc function is used for this purpose:
- E. Lorch in J. Phys. C in 1969. Actually, he was going the opposite way – from I(Q) to g(r), but the truncation effect is similar,
- G. Gutiérrez et al. in PRB in 2002 (copy) – the “window function” in eq. (2) there,
- Z. Lin & L. Zhigilei in PRB in 2006 (copy) – the “damping function”, eq. (8) there.
The continuous density approximation from the previous section, in histogram approximation, replaced n(r) in eq. (1) with
With the sinc damping (option --sinc
in Debyer) n(r) is replaced by:
Computational approach¶
Scattering factors¶
Atomic scattering factor are assigned automatically, by searching atom names in built-in tables. These tables are based on
- International Tables for Crystallography, Volume C 1992, Table 6.1.1.4 (pp. 500-502), Coefficients for analytical approximation to the scattering factors (for x-rays)
- Neutron scattering lengths and cross-sections (for neutrons)
Histogram approximation¶
Distance-histogram approximation avoids calculation of expensive sine function for each atomic pair. It splits computations into two steps.
- calculate a histogram of distances for each pair of atomic species (the most computationally intensive step),
- the Debye formula is applied treating distances in single histogram bin together.
For monoatomic system it can be written as:
where nk and rk are the number of pairs and the distance corresponding to the k-th bin.
The errors from this approximation (when using default histogram bin size) are negligible.
(TODO: describe not used alternatives, such as spherical harmonics approx.)
Cell lists¶
When cut-off rc is small cell lists are used to speed up computations (currently it’s implemented only for systems in PBC).
Parallelization¶
For large systems, almost all the processor time is spent calculating atomic distances. Only this part is parallelized and optimized. The program can be built either as serial or parallel for running on a single computer (with OpenMP) or parallel for running on a cluster (with MPI). See the README file for details how to build it.
Note that compiler options related to floating point arithmetics
may notably improve performance. Debyer is safe to compile with
the -ffast-math
(or equivalent) compiler option.
Computing all the distances between atoms scales with N^2 (N=number of atoms). It scales linearly when using constant cut-off for interatomic distances.
To give a feeling how it scales with the system size, here are rough estimates for calculating all distances in a SiC grain, on two cores of a 2013 computer:
- grain size 5nm – 10,000 atoms – it’s quick
- 10nm – 100,000 atoms – under a minute
- 20nm – 1M atoms – an hour
- 50nm – 10M atoms – days, better to run it in parallel on a cluster
- 100nm – 100M atoms – probably not worth it
Usage¶
debyer --help
shows this summary:
Usage: debyer [OPTIONS]... [FILE]...
-h, --help Print help and exit
--full-help Print help, including hidden options, and exit
-V, --version Print version and exit
-q, --quiet silent mode
-v, --verbose increase verbosity level (can be used twice)
--bench=ATOM-COUNT benchmark - reports time of ID calculation for
ATOM-COUNT atoms.
Stage 1: calculation of ID (Interatomic Distances):
-r, --cutoff=FLOAT cut-off distance for ID calculation
--quanta=FLOAT ID discretization quanta (default=`0.001')
-a, --pbc-a=FLOAT PBC box length in x direction
-b, --pbc-b=FLOAT PBC box length in y direction
-c, --pbc-c=FLOAT PBC box length in z direction
--sample=INT calculate ID by random sampling n atoms
-d, --save-id[=FILENAME] save ID to file
Stage 2: calculation of the final result from ID:
Group: mode
what to calculate - pick one
-x, --xray x-ray powder pattern
-n, --neutron neutron powder pattern
-S, --sf scattering factor (a.k.a total scattering
structure function)
-R, --RDF radial distribution function (RDF)
-g, --PDF pair distribution function (PDF)
-G, --rPDF reduced PDF
Output range (for example -f20 -t100 -s0.1):
-f, --from=FLOAT start of calculated pattern
-t, --to=FLOAT end of calculated pattern
-s, --step=FLOAT step of calculated pattern
Options for real space patterns:
-w, --weight=STRING weighting by x-ray f(0) or neutron b (possible
values="x", "n", "1" default=`1')
-p, --partials include partials as extra columns
Options for reciprocal space patterns:
-l, --lambda=FLOAT wavelength (omit for a pattern in Q)
--sinc apply sinc damping (requires cut-off)
Options valid for both real and reciprocal space patterns:
--ro=FLOAT numeric density, required for (r)PDF and
diffractogram with cutoff
-o, --output=FILENAME output filename. If not given, will be
auto-generated.
IDs (interatomic distances) can be saved with -d
.
It is useful only when calculating multiple patterns of the same sample.
The saved ID can be then used as an input file.
Supported file formats: AtomEye extended CFG, DL_POLY CONFIG/REVCON, LAMMPS input file, XMOL XYZ, and plain text with x y z symbol or symbol x y z in each line. See dbr_conv for details.
PBC:
if the input file contains the box size (e.g., in AtomEye CFG, LAMMPS and
DL_POLY file formats) and the options -a
, -b
, -c
are not given,
the size from the file is used. This is really handy – if you have PBCs
it is better to avoid the xyz format which does not store the PBC cell.
To discard PBC use options -a0
, -b0
, -c0
.
Only orthorhombic PBCs and only PBCs in all 3 dimensions are supported.
Sampling large systems: when the --sample=N
option
is specified, N atoms are randomly chosen and only atomic distance
between these N atoms and all other atoms are calculated.
The only point of this option is to make computations faster (but less exact).
Units of length should be used consistently. The program is unit agnostic.
If the coordinates in the input file are in Angstroms,
the --lambda
option should be also in Å,
the values of \(Q\) will be in Å-1, and so on.
Other units: If the wavelength is specified, diffraction angle in the output is in degrees of \(2\theta\) (otherwise it is \(Q\)).
Examples¶
Calculate diffraction pattern calculation:
debyer -x -f1 -t20 -s0.01 -l0.1 -o t.dat zns.xyz
TBC
Companions¶
Debyer is distributed with a few other tools that were written for very specific tasks and may not be ever re-used – but just in case, I describe them briefly here.
dbr_extend¶
Various modifications of atomistic system in orthorhombic PBC.
This program was written with bicrystal geometry in mind.
Options --help
and --show-examples
display basic documentation.
$ dbr_extend --help
dbr_extend 0.3
Usage: dbr_extend [OPTIONS]... [FILE]...
-h, --help Print help and exit
-V, --version Print version and exit
--show-examples show examples, can be more useful than --help
-v, --verbose increase verbosity level (can be used 3x)
-s, --min-cell=FLOAT (internal) size (lower limit) of cells used for
searching atoms [A] (default=`2.0')
-r, --reduced use reduced coordinates (between 0 and 1)
Most of the actions in this program requires a defined `slab'.
The slab is constructed by a bounding plane (which must be normal
to one of the x, y, z axes) and width.
-x, --x=FLOAT defines a plane x=FLOAT
-y, --y=FLOAT defines a plane y=FLOAT
-z, --z=FLOAT defines a plane z=FLOAT
-b, --bound=ENUM for system finite in given direction: set a
plane to the lower (x,y,z) or upper (X,Y,Z)
bound (possible values="x", "X", "y",
"Y", "z", "Z")
-w, --width=FLOAT width of the slab that will be processed (or
|t|)
If width is not given, it will be equal to the shortest translation vector
found by the program. The following options affect searching of this vector.
-e, --epsilon=FLOAT epsilon used to compare coordinates
(default=`0.1')
-m, --min-delta=FLOAT min length of translation vector
-M, --max-delta=FLOAT max length of translation vector
-p, --periodic require translational symmetry of all the
system
Actions to be performed on atomic configuration file.
-d, --delete delete atoms in the slab
-C, --cut cut out the slab and change PBC accordingly
-u, --add-vacuum add vacuum 'slab'; extends PBC
-c, --add-copy[=N] multiplicate the slab N times (extend PBC and
add atoms) (default=`1')
-a, --add=WIDTH extend PBC by WIDTH and fill the new space
using translation symmetry found in the
defined slab
-N, --multiply=NxNxN multiply configuration in x y and z
-S, --shift=x,y,z shift the system (all atoms) under PBC
-R, --resize=x,y,z resize the system, changing interatomic
distances; the argument is either new size (0
means no change) or another filename
-F, --find-trans find translation symmetries
-U, --make-cubic[=a1,a2,a3,b1,b2,b3,c1,c2,c3]
remove small distortions from perfect cubic
lattice
--merge merge atoms that are closer to each other than
epsilon
--t1 transformation defined in the code as
tranform1()
-D, --density calculate numeric density of the slab
Output file. Default is a dry run.
-i, --in-place replace input file with output
-o, --output=FILENAME output file
For usage examples, invoke the program with --show-examples option.
> dbr_extend --show-examples
USAGE EXAMPLES
dbr_extend -z41.5 -e0.2 -vvv file.cfg
Try to find periodicity of the structure in the z direction starting from
z0=41.5. It tries to find z1 such that each atom with z0 < z < z1 has
a periodic image with x'=x, y'=y, z'=z+delta, delta=z1-z0.
Comparisons of coordinates are done with epsilon=0.2.
dbr_extend -c10 -z41.5 -e0.2 -i file.cfg
Extend PBC box by in the z direction by 10 times the value reported
from the command above, copy atoms to the newly created space,
write the configuration back to the same file.
dbr_extend -bz -w3 -d -i file.cfg
dbr_extend -bZ -w-3 -d -i file.cfg
(Initially, file.cfg contained a slab with surfaces normal to z axis.)
Delete surfaces (3A deep) of the slab.
dbr_extend -v -bz -w2 -a3 -o tmp2.cfg tmp1.cfg
dbr_extend -v -bZ -w-2 -a3 -o tmp3.cfg tmp2.cfg
(Initially, tmp1.cfg contained a slab with surfaces normal to z axis.)
Extend the slab in the z direction, 3A from each surface.
dbr_extend -S0,0,0.5 -r -i file.cfg
Shift object under PBC, by half of the PBC box, in the z direction,
write the configuration back to file.cfg.
dbr_extend -N1x2x1 -o out.cfg file.cfg
Duplicate the system in the y direction (create a supercell).
dbr_extend -v -z-7 -w12. --density file.cfg
Calculate (in a smart way) numeric density of the slab defined by planes
z=-7 and z=5 (its in PBC, so it's continuus region that includes z=0).
Designed to calculate density of a GB in bicrystal geometry.
dbr_extend --resize=ref.cfg -o output.cfg input.cfg
Resize the PBC box, make it the same as the size of the file ref.cfg.
Atomic positions are scaled with the box.
dbr_conv¶
dbr_conv converts between atomistic model file formats. It supports:
- AtomEye extended CFG,
- DL_POLY CONFIG/REVCON,
- LAMMPS data format,
- XMOL XYZ,
- PDB (write only)
- plain format (xyza): each line contains either x y z symbol or symbol x y z.
Compressed files (.gz, .bz2) can be read, but not written.
DL_POLY supports polarizable shell models and the CONFIG/REVCON format contains positions of both atoms and shells. Debyer relies on the convention that shells have names ending with one of the following strings: _sh, _shl, _shel, _shell, -sh, -shl, -shel, -shell. The shells are ignored when reading files.
LAMMPS data file format does not contain contain atom types, only numbers
that are associated to types in an input script to LAMMPS.
Here we use a convention (both when writing and reading the file)
that atom types are given in a comment after the “atom types” line, e.g.
2 atom types # C Si
means that type 1 is C and type 2 is Si.
Another program for converting atomistic file formats is mdfile.py from gosam (Python).
$ ./dbr_conv -h
Usage: dbr_conv [OPTIONS]... INPUT_FILE OUTPUT_FILE
dbr_conv [OPTIONS...] -t TO_FORMAT -m INPUT_FILE1 [INPUT_FILE2]...
-h Print help and exit
-V Print version and exit
-q Silent mode
-v Increase verbosity level (can be used twice)
-s Sort atoms by atomic symbols.
-f FORMAT Convert file from format.
-t FORMAT Convert file to format.
-m Convert multiple files. Output files have only file extension
changed.
Supported format names: atomeye, dlpoly, lammps, pdb, xyz, xyza.
Compressed files (.gz, .bz2) can be read, but not written.
PDB format can only be written.
others¶
- dbr_bonds – show some statistics about bonds in the system.
The program takes two arguments: input file and maximum bond length, e.g:
dbr_bonds file.cfg 2.1
. - dbr_sic – utility to calculate coordination numbers of atoms, so-called ring distribution and other features of zinc-blende structure. Named ‘sic’ because it was used to study SiC structure.
gosam¶
It is actually a separate set of programs, available at https://github.com/wojdyr/gosam/ .
Gosam is a bunch of Python scripts that can:
- create monocrystal in PBC box (atomistic model),
- create bicrystals with coincidence site lattice (CSL) boundaries in PBC box,
- create crystalline grains of given shape, with vacancies, thermal vibrations, etc.
- read/write several file formats (AtomEye cfg, VASP POSCAR, LAMMPS, DL_POLY, XMOL XYZ).