Three-Dimensional Statistical Data Security Problems

  • Authors:
  • Robert W. Irving;Mark R. Jerrum

  • Affiliations:
  • -;-

  • Venue:
  • SIAM Journal on Computing
  • Year:
  • 1994

Quantified Score

Hi-index 0.00

Visualization

Abstract

Suppose there is a three-dimensional table of cross-tabulated nonnegative integer statistics, and suppose that all of the row, column, and "file" sums are revealed together with the values in some of the individual cells in the table. The question arises as to whether, as a consequence, the values contained in some of the other (suppressed) cells can be deduced from the information revealed. The corresponding problem in two dimensions has been comprehensively studied by Gusfield [SIAM J. Comput., 17 (1988), pp. 552--571], who derived elegant polynomial-time algorithms for the identification of any such "compromised" cells, and for calculating the tightest bounds on the values contained in all cells that follow from the information revealed. In this note it is shown, by contrast, that the three-dimensional version of the problem is NP-complete. It is also shown that if the suggested row, column, and file sums for an unknown three-dimensional table are given, with or without the values in some of the cells, the problem of determining whether there exists any table with the given sums is NP-complete. In the course of proving these results, the NP-completeness of some constrained Latin square construction problems, which are of some interest in their own right, is established.