Compression of Sparse Matrices

We consider a simple method to compress a sparse (nn) random matrix into a table of size O(n) which works in expected linear time O(n). The worst case time to access entries in the compressed table is O(1). The compression scheme is based on a random greedy algorithm which places every row of a matrix with n nonzero entries into a table of size n. This minimal table size is achieved with high probability. In case of failure, the table is extended. Experimental results show that the algorithm is very eecient, even for small inputs.