As an implementation optimization, certain axes in the binary tables may be omitted.^[3] These omissions provide a means of data compression, and do not affect the set of axes over which the data apply. This form of data compression is not done “on-the-fly”, but is instead configured before the start of an integration, and merely constrains the data that is written to the data stream. Logically, the data in each data table are always applicable over all axes, whereas the tables in the data stream, created with the knowledge that data will not vary over some of the axes, may be implemented with a reduced dimensionality. This optimization is an EVLA extension, which has no counterpart in the ALMA format at this time. An omitted axis indicates simply that the data are the same for all values along that logical axis (in the subspace defined by the coordinate values on the (row-major-wise) preceding axes.) The designation of omitted axes is provided in the data stream header. See the section called “Example of omitted axes in binary table” for an example of a binary table using omitted axes.

Allowing table axes to be omitted has another application besides data compression. All binary tables may vary over all seven axes, but, in some instances, certain axes will never appear in practice. Thus, some tables will appear to be, for example, six dimensional rather than seven dimensional. However, under our model, tables with reduced dimensionality may be implemented as seven dimensional tables with omitted axes. The only difference in this application compared with the use of omitted axes for data compression is that the logical table index may be presented as having fewer than seven dimensions. Because this application of axes omission requires changes to neither the implementation nor the specification of the data format, we will continue to present the specification as if all binary data tables had seven axes, with the allowance that axes may be omitted.