There are specific tools that make searching this database easier:

la sutysisku dictionary, an offline enabled web app

la vlasisku search engine for the Lojban dictionary

        Word: daigno [jbovlaste]
        Type: fu'ivla
  Gloss Word: counterdiagonal in the sense of "matrix"
  Gloss Word: diagonal in the sense of "matrix"
  Gloss Word: sampling of matrix/tensor entries in the sense of "exactly one from each of the specified rows, columns, etc."
  Gloss Word: subdiagonal in the sense of "matrix"
  Gloss Word: superdiagonal in the sense of "matrix"
  Definition: x1 (ordered list) is a sampling of entries of matrix/tensor x2
       in which exactly one entry is sampled from each row and/or
       column (etc.) between entries x3 (list; default: the largest
       'square'/'hypercubic' sampling possible in the entire tensor
       starting with the first entry, see notes) inclusively following
       selection procedure/rule/function/order x4 (default:
       diagonally, see notes), where the tensor/matrix is expressed in
       basis/under conditions x5
       Notes: Entries of the list in x3 need not actually be sampled; the
       entries listed are merely to name the minimal and maximal
       indices between which the sampling may be drawn.  Thus, the
       indices/labels specified are included in the range of sampling;
       id est: if the matrix entries listed belong to the ith row and
       jth column and the (i+n)th row and (j+m)th column respectively
       (for positive integers i,j,n,m), then the sampling will be
       conducted in all rows of number between (and including) i and
       i+n (yielding n+1 sampled rows) and in all columns of number
       between (and including) j and j+m (yielding m+1 sampled
       columns). The default diagonal sampling procedure for x4 is as
       follows: The first sampled entry has the minimum allowed (as
       specified in x3) indices.  All latter sampled entries (by
       default) have indices of the immediately previous sampled entry
       each augmented by 1. (Which is to say that if the kth sampled
       entry has indices (x,y,...), in that order, then the (k+1)th
       sampled entry has indices (x+1,y+1,...), in that order and
       where each subsequent index would be the respective index of
       the kth sampled entry augmented by 1).  The process terminates
       generally whenever exactly one entry is sampled from each of
       the rows, each of the columns, etc. of the tensor. In the
       default, the process terminates when at least one of the
       indices of a sampled entry of the tensor is as large as
       possible in the range specified by x3. Thus, in order to
       reconcile the general and the default termination conditions,
       the range specified by x3 must be compatible with both; id est:
       it must be a r-dimensional hypercube of entries, so to speak,
       where r is the rank of tensor x2. The default for sampling
       range x3 is between and including the entry in the first row
       and first column (etc.) and the entry in the last row and last
       column (etc.) for an r-dimensional hypercube tensor (meaning
       that each row, column, etc. of the tensor has exactly the same
       number of entries as the others).  Generally, the default range
       begins with the entry of indices each minimal in the tensor
       (called 'the first entry') and extends to include ("draw") the
       maximal r-dimensional hypercube of entries in the tensor with
       one vertex on the first entry; in other words, if the minimum
       of the set of maximal indices in the tensor is g, then the
       sampling range is every row between the first and the gth,
       every column between the first and the gth, etc.  Generally,
       the sampling range must be an r-dimensional orthotope of some
       positive size (that is to say: including at least one entry) no
       larger than the tensor itself, but with the freedom to place at
       most r of its vertices among the entries thereof; if the
       default sampling procedure x4 is being used, then the
       r-dimensional orthotope must be an r-dimensional hypercube.
       Generalizes to any tensor, but is only interesting for tensors
       of rank at least 1. Any mention of geometric terminology (such
       as mention of diagonals, orthotopes, etc.) in the definition or
       notes of this word should be interpreted cautiously and is not
       necessarily good Lojbanic practice; such terminology should not
       necessarily be emulated in practicing Lojbanic thought or
       speech. Not for use for geometric diagonals (such as between
       vertices); confer: digno.