Help With Submit Form

The following is a list of the fields present on the Matrix Market Submit Forms, along with a brief explanation of their purpose.

Locations can be specified in one of the following ways:

• URL http address : (e.g. http://www.cs.univ.edu/~myname/dir) May be either a single matrix or a directory of related matrices.

• URL ftp address: (e.g. ftp://ftp.cs.univ.edu/pub/dir) May be either a single matrix or a directory of related matrices.

• (leave blank) : if there is no public access for this. We will contact you via email to make further arrangements.

If there are any problems encountered while processing your submission we need to have a way of contacting you. Furthermore, without any record of who submitted a matrix, it makes it difficult to update and maintain these database entries. Finally, we need to verify it is indeed you who is submitting these (and not just someone typing in your name). It's not foolproof, and if this becomes a problem, we will start asking for PGP signatures and the like...

If you don't have an email account, and are using a friend's browser, enter a phone number or some other way of contacting you.

This indicates the general linear algebraic problem for which the matrix was originally generated. The choices are as follows. (In this discussion, A and B denote matrices, x a vector, and z a scalar.)

• Linear System ... The solution of systems of linear equations.
• Eigenvalue ... The computation of eigenvalues and/or eigenvectors of matrices. Both simple eigenproblems (Ax = zx)and generalized eigenproblems (e.g., Ax = zBx) are included.
• Least Squares ... The solution to overdetermined systems of linear equations in the least squares sense, that is, minimize || Ax-b || over x where ||.|| denotes the Euclidian norm.
• Structure ... Studies of the nonzero structure of sparse matrices under various transformations.

Included in each of these problems are the study of matrix factorizations peculiar to those problems.

Here you can provide a documentation or README file that provides background information about the matrix data, such as what application it originates from, or any other interesting characteristics.

For a documentation example of current Matrix Market submissions, see one of the Harwell-Boeing entries, such as the ACOUST set.

Although one can type in documentation directly into an HTML form, (such as the "References" slot below), the formatting of the text is lost. This makes it difficult, for example, to include tables or Fortran comments.

Include references related to either the matrix data or the application which the matrix data is related to. BibTeX formats are preferred.

This is used to informally characterize the nonzero structure of the matrix. Note that a given matrix may have more than one of these properties. Additional properties will be added to this list as the Matrix Market collection grows. The choices are as follows.

• Sparse ... A matrix which contains only a few nonzero elements in each row or column.
• Banded ... A sparse matrix in which all the nonzeros are clustered about the main diagonal. The bandwidth is the largest |i-j| such that a(i,j) is nonzero.
• Diagonal ... A matrix of bandwidth 0, i.e., its only nonzeros lie along the main diagonal.
• Tridiagonal ... A matrix of bandwidth 1, i.e., only the elements a(i,i-1), a(i,i), a(i,i+1) can be nonzero in row i.
• Block Tridiagonal ... A matrix which can be partitioned into blocks such that the only blocks containing nonzeros in each block row are the diagonal block, its predecessor, and its successor. Finite difference discretizations of 2D elliptic operators typically give rise to block tridiagonal matrices in which each of the nonzero blocks are themselves tridiagonal.

This indicates the definiteness properties of the matrix. The choices are as follows.

• Positive Definite.
• Positive Semidefinite.
• Negative Definite.
• Negative Semidefinite.
• Indefinite.

A matrix A is positive definite if and only if x'Ax > 0 for all nonzero vectors x, where ' denotes transpose. A symmetric positive definite matrix has all positive real eigenvalues. A is negative definite if and only if -A is positive definite. For semidefiniteness, the inequality above is relaxed to admit equality; a semidefinite matrix has at least one zero eigenvalue. An indefinite matrix has none of these special properties.

[ Home ] [ Search ] [ Browse ] [ Resources ] [ Submit ]

Last change in this page : January 11, 1997 by RFB.