This is a basic introduction to this concept. For a more mathematical take, see dual list.
A comma basis is a basis for the null-space (sometimes also called the "kernel") of a temperament. It consists of [math]n[/math] linearly independent vectors, where [math]n[/math] is the nullity, each one representing a comma that is tempered out.
Linear independence means that no comma can be found as the sum of any multiples of the other commas. For example, consider the set of three commas 81/80, 126/125, and 225/224. As vectors those are [-4 4 -1 0⟩, [1 2 -3 1⟩, and [-5 2 2 -1⟩. Notice that the third comma is actually the difference between the other two; [-4 4 -1 0⟩ - [1 2 -3 1⟩ = [-5 2 2 -1⟩, or as cents, 21.51¢ - 13.80¢ = 7.71¢. So, if two of these three commas are ever tempered out (mapped to 0¢), then the third one necessarily is also tempered out. Therefore, we only need to pick two of these commas to put in our comma basis; the third one would be implied.
The comma basis can be thought of either as a list of vectors or as a matrix formed by putting these vectors (as columns) together. Besides, it is often presented in terms of ratios for convenience. Various normal forms have been developed as identifiers of temperaments.
With respect to the mapping
The comma basis is considered the dual of the temperament's mapping matrix. Temperaments may be identified by either their mapping or comma basis.
Functions for finding the null-space of a matrix are readily available in many math libraries. All you need to do to get a comma basis for a mapping is to find the null-space. To learn more about finding the null-space, see Douglas Blumeyer's RTT How-To #The other side of duality.
To reverse the null-space operation, that is, to find a mapping from a comma basis, you can also use the null-space operation; the relationship between a matrix and its null-space essentially works both ways.
Some math libraries, such as Sage, provide functions for both directions; in Sage, to go from a mapping to a comma basis, use
left_kernel(), and to go from a comma basis to a mapping, use
right_kernel(). In other math libraries, such as Wolfram Language, the null-space operation
NullSpace is primarily designed to work for mappings, and so if you want correct results, you must transform the basis for the null-space into a mapping-like form, perform the null-space operation, and then undo the initial transformation.
This transformation that relates the two directions of null-space operations is called the transpose. It works by reflecting a matrix's values across its main diagonal, i.e. either the diagonal running from the top-left corner toward the bottom-right, or the diagonal running from the bottom-right corner toward the top-left.
However, transposing a comma basis, using a mapping-style null-space function, then transposing again, will return a mapping in a strange form, with all of its zeros in the top-right corner, rather than the bottom-left as is preferred. The solution for this problem is to use the anti-transpose instead of the transpose. This is the same you reflect the matrix's entries across its anti-diagonal (starting from either the top-right or bottom-left corner).
You can remember this because most mappings and comma bases have zeroes in the bottom-left corner, and you want to keep them there; some kind of transpose is necessary to convert the constituent comma vectors columns of the comma basis into rows as if they were constituent generator maps (rows) of a mapping, but a normal transpose of the comma basis would flip its zeroes into the top-right corner instead.