indexer-choice.rst

Which indexing method(s) should I use?

The short answer

For most applications, the best choice of indexing method is xgandalf.

In most cases, it's not necessary to use several algorithms in sequence (e.g. --indexing=xgandalf,mosflm,asdf).

Read on for recommendations in specific cases.

Unknown crystal lattice parameters

Most of the indexing algorithms run fine without prior information about the lattice parameters. The indexing algorithms will generate a set of estimated lattice parameters for each crystal, and you can use cell_explorer to plot histograms to determine the most "popular" parameters.

Problems arise because there are an infinite number of ways to represent any given lattice. Different indexing algorithms have different "opinions" about which representation to use when not constrained by prior information.

In this case, mosflm is a good choice of indexing method because it can detect centered lattices instead of producing a primitive unit cell every time. This makes it easier and quicker to see the correct symmetry (including indexing ambiguities) and get to the crystallographically conventional representation of the structure.

Other indexing methods, particularly xgandalf, are fine as long as you use proper crystallographic knowledge to determine the symmetry for merging and structure solution.

It's particularly advisable to avoid using multiple indexing methods together when the lattice parameters are unknown, because the "preferences" of different algorithms will be mixed up together.

Fastest indexing (e.g. real-time analysis pipelines)

Read the document about data processing speed. If you have a GPU available, use ffbidx. Otherwise, the best choice is probably asdf or xgandalf with --xgandalf-fast, but there are many other considerations for increasing processing speed.

Wide-bandwidth X-ray diffraction (Laue/pink beam)

Use pinkindexer, especially if the (FWHM) bandwidth is larger than about 3% of the wavelength. For smaller bandwidths, try with xgandalf.

For larger bandwidths, it's also advisable to use options --no-refine and --no-check-peaks, because these parts of the program assume that the bandwidth is small.

Electron diffraction data

Use pinkindexer, but first read the document about electrons.

Small unit cell

For very small unit cells (<20 Angstrom axis length), use smallcell. For larger (but still "small") cells, researchers have also had success using xgandalf, and it's also worth trying with taketwo.

Multiple overlapping diffraction patterns

Try felix, which is specifically designed for heavily overlapping diffraction patterns. Note, however, that it's unfortunately difficult to get hold of a copy of the Felix executable.

CrystFEL was installed minimally

The taketwo and smallcell algorithms have no dependencies, neither compile-time nor run-time, so are both always available. Note however that both always need prior lattice parameters, and smallcell is a specialist algorithm for small unit cells (i.e. not macromolecules).

Even if you have no system-wide access to the computer system, you can still make a user-local installation of dirax, mosflm, xds or felix. These methods have no compile-time requirements, and only need the corresponding programs to be available at run-time.