*MOPER
*MOPER,
ParR
, Par1
,
Oper
, Val1
,
Val2
, Val3
,
Val4
, Val5
,
Val6
Performs matrix operations on array parameter matrices.
Argument Descriptions
ParR
The name of the resulting array parameter matrix. See *SET for name restrictions.
Par1
First array parameter matrix input to the operation.
Oper
Matrix operations. Usage of the
Val1
throughVal6
arguments varies for each operation, as described below:- INVERT --
(*MOPER,
ParR
,Par1
, INVERT)Square matrix invert: Inverts the n x n matrix in Par1 into ParR. The matrix must be well conditioned.
Warning: Non-independent or ill-conditioned equations can cause erroneous results.
For large matrices, use the APDL Math operation *LSFACTOR for efficiency (see APDL Math).
- MULT --
(*MOPER,
ParR
,Par1
, MULT,Par2
)Matrix multiply: Multiplies
Par1
byPar2
. The number of rows ofPar2
must equal the number of columns ofPar1
for the operation. IfPar2
is input with a number of rows greater than the number of columns ofPar1
, matrices are still multiplied. However, the operation only uses a number of rows ofPar2
equal to the number of columns ofPar1
.- COVAR --
(*MOPER,
ParR
,Par1
, COVAR,Par2
)Covariance: The measure of association between columns of the
Par1
input matrix.Par1
of size m runs (rows) by n data (columns) is first processed to produce a row vector containing the mean of each column, which is transposed to the output column vectorPar2
of n array elements. ThePar1
andPar2
operation then producesParR
, a resulting n x n matrix of covariances (with the variances as the diagonal terms).- CORR --
(*MOPER,
ParR
,Par1
, CORR,Par2
)Correlation: The correlation coefficient between columns of the
Par1
input matrix.Par1
of size m runs (rows) by n data (columns) is first processed to produce a row vector containing the mean of each column, which is then transposed to the output column vectorPar2
of n array elements. ThePar1
andPar2
operation then producesParR
, a resulting n x n matrix of correlation coefficients (with a value of 1.0 for the diagonal terms).- SOLV --
(*MOPER,
ParR
,Par1
, SOLV,Par2
)Solution of simultaneous equations: Solves the set of
n
equations ofn
terms of the form an1x1 + an2x2 + ... + annxn = bn wherePar1
contains the matrix of a-coefficients,Par2
contains the vector(s) of b-values, andParR
contains the vector(s) of x-results.Par1
must be a square matrix. The equations must be linear, independent, and well conditioned.Warning: Non-independent or ill-conditioned equations can cause erroneous results.
For large matrices, use the APDL Math operation *LSFACTOR for efficiency (see APDL Math).
- SORT --
(*MOPER,
ParR
,Par1
, SORT,Par2
,n1
,n2
,n3
)Matrix sort: Sorts the columns of matrix
Par1
according to sort vectorPar2
and places the result back intoPar1
. Rows ofPar1
are moved to the corresponding positions indicated by the values ofPar2
.Par2
may be a column ofPar1
(in which case it will also be reordered). Alternatively, you may specify the column ofPar1
to sort usingn1
(leavingPar2
blank). A secondary sort can be specified by columnn2
, and a third sort using columnn3
.ParR
is the vector of initial row positions (the permutation vector). SortingPar1
according toParR
should reproduce the initial ordering.- NNEAR --
(*MOPER,
ParR
,Par1
, NNEAR,Toler
)Nearest Node: Finds the nodes nearest to the given set of points in
Par1
, wherePar1
is an n x 3 array of coordinate locations.ParR
is a vector of the nearest selected nodes, or 0 if no nodes are nearer thanToler
.Toler
defaults to 1 and is limited to the maximum model size.- ENEAR --
(*MOPER,
ParR
,Par1
, ENEAR,Toler
)Nearest Element: Finds the elements nearest to the given set of points in
Par1
, wherePar1
is an n x 3 array of coordinate locations.ParR
is a vector of the nearest selected elements, or 0 if no element centroids are nearer thanToler
.Toler
defaults to 1 and is limited to the maximum model size.- MAP --
(*MOPER,
ParR
,Par1
, MAP,Par2
,Par3
,kDim
,--
,kOut
,LIMIT
)Maps the results from one set of points to another. For example, you can map pressures from a CFD analysis onto your model for a structural analysis.
Par1
is the Nout x 3 array of points that will be mapped to.Par2
is the Nin x M array that contains M values of data to be interpolated at each point and corresponds to the Nin x 3 points inPar3
. The resultingParR
is the Nout x M array of mapped data points.For each point in the destination mesh, all possible triangles in the source mesh are searched to find the best triangle containing each point. It then does a linear interpolation inside this triangle. You should carefully specify your interpolation method and search criteria in order to provide faster and more accurate results (see
LIMIT
, below).kDim
is the interpolation criteria. IfkDim
= 2 or 0, two dimensional interpolation is applied (interpolate on a surface). IfkDim
= 3, three dimensional interpolation is applied (interpolate on a volume).kOut
specified how points outside of the domain are handled. IfkOut
= 0, use the value(s) of the nearest region point for points outside of the region. IfkOut
= 1, set results outside of the region to zero.LIMIT
specifies the number of nearby points considered for interpolation. The default is 20, and the minimum is 5. Lower values will reduce processing time; however, some distorted or irregular sets of points will require a higherLIMIT
value to encounter three nodes for triangulation.Output points are incorrect if they are not within the domain (area or volume) defined by the specified input points. Also, calculations for out-of-bound points require much more processing time than do points that are within bounds. Results mapping is available from the command line only.
- INTP --
(*MOPER,
ParR
,Par1
, INTP,Par2
)Finds the elements that contain each point in the array of n x 3 points in
Par1
.Par2
will contain the set of element ID numbers andParR
will contain their n x 3 set of natural element coordinates (values between -1 and 1).Par1
must be in global Cartesian coordinates.- SGET --
(*MOPER,
ParR
,Par1
, SGET,Par2
,Label
,Comp
)Gets the nodal solution item corresponding to
Label
andComp
(see the PLNSOL command) and interpolates it to the given element locations.Par1
contains the n x 3 array of natural element coordinates (values between -1 and 1) of the n element ID numbers inPar2
.Par1
andPar2
are usually the output of the *MOPER,,,INTP operation.ParR
contains the n interpolated results.
Val1
,Val2
,...
,Val6
Additional input used in the operation. The meanings of
Val1
throughVal6
vary depending on the specified matrix operation. See the description ofOper
for details.
Notes
Each starting array element number must be defined for each array parameter matrix if it does not start at the first location. For example, *MOPER,A(2,3),B(1,4),MULT,C(1,5) multiplies submatrix B (starting at element (1,4)) by submatrix C (starting at element (1,5)) and puts the result in matrix A (starting at element (2,3)).
The diagonal corner elements for each submatrix must be defined: the upper left corner by the
array starting element (on this command), the lower right corner by the current values from
the *VCOL and *VLEN commands. The default values are the
(1,1) element and the last element in the matrix. No operations progress across matrix planes
(in the 3rd dimension). Absolute values and scale factors may be applied to all parameters
(*VABS, *VFACT). Results may be cumulative
(*VCUM). Array elements should not be skipped with the
*VMASK and the NINC
value of the
*VLEN specifications. See the *VOPER command for
details.
This command is valid in any processor.