c155.purged.gz file contains the matrix output by the purge step
(input of the merge step), while the other files correspond to the matrix
output by the merge/replay steps (input of the linear algebra step).
The c155.purged.gz file contains a matrix defined over GF(2),
with 17117966 rows and
17117806 columns, where the column indices go from 0 to 56385930.
After a header line containing that information, each row is encoded by one
line, with a,b: followed by a list of hexadecimal numbers
from 0 to 56385930 for the column indices for that row.
The matrix represented by the other files is also defined over GF(2), it has 3854889 rows and 3854728 columns, with 170 non-zero coefficients per row on average. It is sorted in binary as a sequence of rows. There is no header. Each row holds a 32-bit little endian integer for the row weight, followed by this very number of 32-bit little endian integers, indicating column indices of non-zero elements.
As is relatively usual with the Number Field Sieve factorization
algorithm, the densest columns of the relation matrix produced by the
algorithm are put aside before linear algebra processing, and the
required correction is performed later on. Here, 32 columns were
put aside, in the c155.dense.bin file.
The row weight and column weight files, for each matrix, are stored
simply as list of 32-bit little endian integers.
| 1.0G | 17117966x17117806, where the column indices go from 0 to 56385930 | ||
| 2.6G | 3854889x3854728, 655331417 non-zero coefficients | ||
| 15M | row weights | ||
| 15M | column weights | ||
| 108M | 3854889x32, 48173007 non-zero coefficients | ||
| 15M | row weights | ||
| <1k | column weights |