17 Outline of the Data Model

 17.1 Components
 17.2 Attributes
 17.3 The connections component and interpolation
 17.4 Position and connection dependent data

The data structures which can be represented using the DX data model include:

(1)
data defined on a regular orthogonal grid,
(2)
data defined on a deformed regular or curvilinear grid,
(3)
data defined on various sorts of irregular grids,
(4)
unstructured data with no connection between the data points.

In the first case the dependent variable (which is the quantity to be visualised), perhaps for example, temperature, is sampled in a regular grid inside some region defined with the independent variables as axes. Usually the independent variables will be positions in two, three or higher dimensions. Typically inside your own programs such data would be represented as an array of appropriate dimensionality. The second and third cases are generalisations of the first and can be ignored for our present purposes. The fourth case corresponds to particle (or catalogue) data. Here each point is simply a point in an assemblage or cloud of points; there is no connection in terms of either the independent variables (positions) or the dependent variable between the points. An example might be a simulation of a globular cluster where each data point corresponds to a separate star following its own orbit inside the cluster.

Any of the four cases can occur with any dimensionality: one, two, three and higher dimensional data can be represented. Similarly all the usual data types are available: integer, real, double precision, complex etc. The dependent quantity may be a scalar (such as temperature, pressure or energy) or a vector (such as velocity or momentum).

The fundamental entity in the DX data model is an object. Data are represented by the same set of objects both when they are resident in memory when DX is running and when they are stored in native format disk files7. There are several sorts of objects. For the present purposes the most important type of object is the field; usually each separate dataset will be represented as a single field8.

A field consists of an arbitrary number of components and each component itself has a number of attributes. This hierarchy is illustrated in Figure 18.


field
  component 1
    attribute 1
    attribute 2
         .
         .
  component 2
    attribute 1
    attribute 2
         .
         .
  component 3
       .
       .

Figure 18: The hierarchy of items in a DX field. Indenting to the right indicates successively lower levels in the hierarchy.


17.1 Components

There are many different types of component. Some are mandatory, but most are optional. Only a few of the more common ones are relevant to the present discussion. They are listed in Table 1 and described below.


Name Mandatory? Comments



positions position of each datum
data value of each datum
connections relation between data points
box bounding box for positions in the field

Table 1: Common components of a field.

Positions
The position component is an array9 defining a position for every datum in the dataset. If the field contains unconnected particle data then the array will be a simple list with a position for each datum. If the field contains a regular grid the positions can be represented more compactly as a regular or product array; essentially just the origin, size and increment of the grid are stored.
Data
The data component is an array storing the dependent variable: temperature, density, velocity, momentum or whatever.
Connections
The connections component prescribes to DX how to perform interpolation between neighbouring grid elements. It is described further in Section 17.3, below. In unstructured particle or catalogue data the connection component is absent (because the position of one particle bears no relation to the position of any other particle; think of the example of stars following their individual orbits within a globular cluster).
Box
The box component is an array of 2n points where n is the dimensionality of the positions component. The values contain the coordinates of a bounding box sufficiently large to just enclose all the positions of a field.

17.2 Attributes

Components can have attributes associated with them. Attributes are additional items of information describing some aspect of the component. For example, the dependency attribute, ‘dep’, identifies another component on which the component depends; the ‘data’ component may depend on either the positions or the connections component, depending on circumstances. Alternatively the component ‘ref’ indicates that a component refers to another component. Typically the connections component will have a ‘ref’ of ‘positions’, indicating that the connections refer to the positions.

17.3 The connections component and interpolation

The connections component prescribes to DX how to perform interpolation between neighbouring grid elements. It describes the logical connectivity between adjacent neighbours in a grid. For example, the interpolation could be performed linearly along one of the axes, or bi- or tri-linearly along two or more axes simultaneously.

The connections component is implemented as an array. Each point in the positions component is given an ordinal number (starting at zero). The connections array comprises a ‘list of lists’ of numbers in which each entry represents the ordinal values of the points that are to be connected, listed in the order that they are to be connected.

You will not normally need to be concerned with the full details of the connections component. For a regular grid it can be created automatically when the data are imported into DX and represented compactly using a path or mesh array.

Fields representing particle or catalogue data where there are no logical connections between elements in the positions component will not usually have a connections component (again think of the example where each point represents a star orbiting in a globular cluster).

17.4 Position and connection dependent data

For a field of gridded data (it is easier to think of a regular grid, though the argument applies equally to an irregular grid) there are two ways in which the dependent variable can relate to the elements of the grid.

Position dependent
The dependent variable represents instantaneous samples of the underlying field (temperature, pressure, momentum or whatever), computed at the position of each grid point.
Connection dependent
The grid points delineate the corners of a cell. The dependent variable is the value inside this cell. An example might be a grid spanning a globular cluster with the dependent variable being the number of stars in each cell.

The practical difference between the two cases is the way in which values for the dependent variable are computed at positions intermediate between grid points.

Position dependent
A value for the dependent variable is interpolated from its values at the neighbouring grid points.
Connection dependent
The value of the dependent variable for the cell is used for any positions inside the cell (because the value pertains to the entire cell, not just one instantaneous point).

The dependency attribute, ‘dep’, of the data component will be set to ‘positions’ if the data are position dependent and to ‘connections’ if they are connection dependent.

Before you input data to DX you must decide whether they are position dependent or connection dependent. Position dependent data are probably more common in astronomy.

7It is also, of course, possible to input files written in a wide variety of other formats; see Sections 5 and 6 for some examples.

8This usage of the term ‘field’ corresponds to its usual meaning in physics rather than in computing. For example, a DX field may represent a three-dimensional grid holding a set of samples of the velocity field throughout some volume. This usage is quite different from the usual meaning in computing: a predetermined section of a record allocated to the storage of a particular data item. Anyone used to the computing terminology should particularly beware of this different usage.

9Strictly speaking this array, and other arrays in components, are themselves DX objects of type array. However, this complication is not important to the present discussion and it is adequate to think of the component as simply an array of numbers.