MATHS

Evaluates mathematical expressions applied to NDF data structures

Description:

This application allows arithmetic and mathematical functions to be applied pixel-by-pixel to a number of NDF  data structures and constants so as to produce a new NDF. The operations to be performed are specified using a Fortran-like mathematical expression. Up to 26 each input NDF data and variance arrays, 26 parameterised ‘constants’, and pixel and data co-ordinates along up to 7 dimensions may be combined in wide variety of ways using this application. The task can also calculate variance estimates for the result when there is at least one input NDF array.

Usage:

maths exp out ia-iz=? va-vz=? fa-fz=? pa-pz=? lbound=? ubound=?

Parameters:

The mathematical expression to be evaluated for each NDF pixel, e.g. "(IA-IB$+$2)$\ast$PX". In this expression, input NDFs are denoted by the variables IA, IB, …IZ, while constants may either be given literally or represented by the variables PA, PB, …PZ. Values for those NDFs and constants which appear in the expression will be requested via the application’s parameter of the same name.

Fortran-77 syntax is used for specifying the expression, which may contain the usual intrinsic functions, plus a few extra ones. An appendix in SUN/61 gives a full description of the syntax used and an up to date list of the functions available. The expression may be up to 132 characters long and is case insensitive.

These parameters supply the values of ‘sub-expressions’ used in the expression EXP. Any of the 26 (FA, FB, …FZ) may appear; there is no restriction on order. These parameters should be used when repeated expressions are present in complex expressions, or to shorten the value of EXP to fit within the 132-character limit. Sub-expressions may contain references to other sub-expressions and constants (PA-PZ). An example of using sub-expressions is:
EXP $>$ PA$\ast$ASIND(FA/PA)$\ast$XA/FA
FA $>$ SQRT(XA$\ast$XA$+$XB$\ast$XB)
PA $>$ 10.1

where the parameter name is to the left of $>$ and its value is to the right of the $>$.

The set of 26 parameters named IA, IB, …IZ is used to obtain the input NDF data structure(s) to which the mathematical expression is to be applied. Only those parameters which actually appear in the expression are used, and their values are obtained in alphabetical order. For instance, if the expression were "SQRT(IB$+$IA)", then the Parameters IA and IB would be used (in this order) to obtain the two input NDF data structures.
Lower bounds of new NDF, if LIKE=! and there is no input NDF referenced in the expression. The number of values required is the number of pixel co-ordinate axes in the expression.
An optional template NDF which, if specified, will be used to define bounds and data type of the new NDF, when the expression does not contain a reference to an NDF. If a null response (!) is given the bounds are obtained via Parameters LBOUND and UBOUND, and the data type through Parameter TYPE. [!]
OUT = NDF (Write)
Output NDF to contain the result of evaluating the expression at each pixel.
The set of 26 parameters named PA, PB, …PZ is used to obtain the numerical values of any parameterised ‘constants’ which appear in the expression being evaluated. Only those parameters which actually appear in the expression are used, and their values are obtained in alphabetical order. For instance, if the expression were "PT$\ast$SIN(IA/PS)", then the Parameters PS and PT (in this order) would be used to obtain numerical values for substitution into the expression at the appropriate points.

These parameters are particularly useful for supplying the values of constants when writing procedures, where the constant may be determined by a command-language variable, or when the constant is stored in a data structure such as a global parameter. In other cases, constants should normally be given literally as part of the expression, as in "IZ$\ast$$\ast$2.77".

Specifies the method by which values for the variance component of the output NDF are calculated. The algorithm used to determine these values involves perturbing each of the input NDF data arrays in turn by an appropriate amount, and then combining the resulting output perturbations. If QUICK is set to TRUE, then each input data array will be perturbed once, in the positive direction only. If QUICK is set to FALSE, then each will be perturbed twice, in the positive and negative directions, and the maximum resultant output perturbation will be used to calculate the output variance. The former approach (the normal default) executes more quickly, but the latter is likely to be more accurate in cases where the function being evaluated is highly non-linear, and/or the errors on the data are large. This parameter is ignored if the expression does not contain a token to at least one input NDF structure. [TRUE]
The title for the output NDF. A null value will cause the title of the (alphabetically) first input NDF to be used instead. [!]
Data type for the new NDF, if LIKE=! and no input NDFs are referenced in the expression. It must be one either "_DOUBLE" or "_REAL".
Upper bounds of new NDF, if LIKE=! and there is no input NDF referenced in the expression. These must not be smaller than the corresponding LBOUND. The number of values required is the number of pixel co-ordinate axes in the expression.
Specifies whether the UNITS component of the (alphabetically) first input NDF or the template NDF will be propagated to the output NDF. By default this component is not propagated since, in most cases, the units of the output data will differ from those of any of the input data structures. In simple cases, however, the units may be unchanged, and this parameter then allows the UNITS component to be preserved. This parameter is ignored if the expression does not contain a token to at least one input NDF structure and LIKE=!. [FALSE]
The set of 26 parameters named VA, VB, …VZ is used to obtain the input NDF variance array(s)  to which the mathematical expression is to be applied. The variance VA corresponds to the data array specified by Parameter IA, and so on. Only those parameters which actually appear in the expression, and do not have their corresponding data-array Parameter IA-IZ present, have their values obtained in alphabetical order. For instance, if the expression were "IB$+$SQRT(VB$+$VA)", then the Parameters VA and IB would be used (in this order) to obtain the two input NDF data structures. The first would use just the variance array, whilst the second would read both data and variance arrays.
Specifies whether values for the VARIANCE component of the output NDF should be calculated. If this parameter is set to TRUE (the normal default), then output variance values will be calculated if any of the input NDFs contain variance information. Any which do not are regarded as having zero variance. Variance calculations will normally be omitted only if none of the input NDFs contain variance information. However, if VARIANCE is set to FALSE, then calculation of output variance values will be disabled under all circumstances, with a consequent saving in execution time. This parameter is ignored if the expression does not contain at least one token to an input NDF structure. [TRUE]

Examples:

maths "ia-1" dat2 ia=dat1
The expression "ia-1" is evaluated to subtract 1 from each pixel of the input NDF referred to as IA, whose values reside in the data structure dat1. The result is written to the NDF structure dat2.
maths "(ia-ib)/ic" ia=data ib=back ic=flat out=result units
The expression "(ia-ib)/ic" is evaluated to remove a background from an image and to divide it by a flat-field. All the images are held in NDF data structures, the input image being obtained from the data structure data, the background image from back and the flat-field from flat. The result is written to the NDF structure result. The data units are unchanged and are therefore propagated to the output NDF.
maths "-2.5$\ast$log10(ii)$+$25.7" ii=file1 out=file2
The expression "-2.5$\ast$log10(ii)$+$25.7" is evaluated to convert intensity measurements into magnitudes, including a zero point. Token II represents the input measurements held in the NDF structure file1. The result is written to the NDF structure file2. If file1 contains variance values, then corresponding variance values will also be calculated for file2.
maths exp="pa$\ast$exp(ia$+$pb)" out=outfile pb=13.7 novariance
The expression "pa$\ast$exp(ia$+$pb)" is evaluated with a value of 13.7 for the constant PB, and output is written to the NDF structure outfile. The input NDF structure to be used for token IA and the value of the other numerical constant PA will be prompted for. NOVARIANCE has been specified so that output variance values will not be calculated.
maths exp="mod(XA,32)$+$mod(XB,64)" out=outfile like=comwest
The expression "mod(XA,32)$+$mod(XB,64)" is evaluated, and output is written to the NDF structure outfile. The output NDF inherits the shape, bounds, and other properties (except the variance) of the NDF called comwest. The data type of outfile is _REAL unless comwest has type _DOUBLE. XA and XB represent the pixel co-ordinates along the x and y axes respectively.
maths "xf$\ast$xf$+$0$\ast$xa" ord2 lbound=[-20,10] ubound=[20,50]
The expression "xf$\ast$xf$+$0$\ast$xa" is evaluated, and output is written to the NDF structure ord2. The output NDF has data type _REAL, is two-dimensional with bounds $-$20:20, 10:50. The XA is needed to indicate that XF represents pixel co-ordinates along the y axis.
maths "xa/max(1,xb)$+$sqrt(va)" ord2 va=fuzz title="Fuzz correction"
The expression "xa/max(1,xb)$+$sqrt(va)" is evaluated, and output is written to the NDF structure ord2. Token VA represents the input variance array held in the NDF structure fuzz. The output NDF inherits the shape, bounds, and other properties of fuzz. The title of ord2 is "Fuzz correction". The data type of ord2 is _REAL unless fuzz has type _DOUBLE. XA and XB represent the pixel co-ordinates along the x and y axes respectively.

Notes:

• The alphabetically first input NDF is regarded as the primary input dataset. NDF components whose values are not changed by this application will be propagated from this NDF to the output. The same propagation rules apply to the LIKE template NDF, except that the output NDF does have inherit any variance information.

• There are additional tokens which can appear in the expression.

The set of seven tokens named CA, CB, …CG is used to obtain the data co-ordinates from the primary input NDF data structure. Any of the seven parameters may appear in the expression. The order defines which axis is which, so for example, "2$\ast$CF$+$CB$\ast$CB" means the first-axis data co-ordinates squared, plus twice the co-ordinates along the second axis. There must be at least one input NDF in the expression to use the CA-CG tokens, and it must have dimensionality of at least the number of CA-CG tokens given.

The set of seven tokens named XA, XB, …XG is used to obtain the pixel co-ordinates from the primary input NDF data structure. Any of the seven parameters may appear in the expression. The order defines which axis is which, so for example, "SQRT(XE)$+$XC" means the first-axis pixel co-ordinates plus the square root of the co-ordinates along the second axis. Here no input NDF need be supplied. In this case the dimensionality of the output NDF is equal to the number of XA-XG tokens in the expression. However, if there is at least one NDF in the expression, there should not be more XA-XG tokens than the dimensionality of the output NDF (given as the intersection of the bounds of the input NDFs).

• If illegal arithmetic operations (e.g. division by zero, or square root of a negative number) are attempted, then a bad pixel will be generated as a result. (However, the infrastructure software that detects this currently does not work on OSF/1 systems, and therefore MATHS will crash in this circumstance.)

• All arithmetic performed by this application is floating point. Single-precision will normally be used, but double-precision will be employed if any of the input NDF arrays has a numeric type of _DOUBLE.

Calculating Variance

The algorithm used to calculate output variance values is general-purpose and will give correct results for any reasonably well-behaved mathematical expression. However, this application as a whole, and the variance calculations in particular, are likely to be less efficient than a more specialised application written knowing the form of the mathematical expression in advance. For simple operations (addition, subtraction, etc.) the use of other applications (ADD, SUB, etc.) is therefore recommended, particularly if variance calculations are required.

The main value of the variance-estimation algorithm used here arises when the expression to be evaluated is too complicated, or too infrequently used, to justify the work of deriving a direct formula for the variance. It is also of value when the data errors are especially large, so that the linear approximation normally used in error analysis breaks down.

There is no variance processing when there are no tokens for input NDF structures.

Timing

If variance calculations are not being performed, then the time taken is approximately proportional to the number of NDF pixels being processed. The execution time also increases with the complexity of the expression being evaluated, depending in the usual way on the nature of any arithmetic operations and intrinsic functions used. If certain parts of the expression will often give rise to illegal operations (resulting in bad pixels), then execution time may be minimised by placing these operations near the beginning of the expression, so that later parts may not need to be evaluated.

If output variance values are being calculated and the QUICK parameter is set to TRUE, then the execution time will be multiplied by an approximate factor ($N+$1), where n is the number of input NDFs which contain a VARIANCE component. If QUICK is set to FALSE, then the execution time will be multiplied by an approximate factor ($2N+$1).