Progress report for UGA

Clément Pernet

Reporting period from March 2017 to June 2018

Finance and administration

Hongguang Zhu, research engineer 100%
- 1 September 2017 – 30 August 2019
- Formerly research assistant at Univ. of Sherbrooke, QC, Canada.
- Working on distributed code for LinBox (D5.14)
Alexis Breust, research engineer 100%
- 1 September 2018 – 30 August 2019
- Previously: video game dev, LinBox research engineer
- Working on distributed code for LinBox (D5.14)

Releases of
- givaro-4.0.3 and givaro-4.0.4:
  - improving build system,
  - autodetection of SIMD
  - micro-instruction,
  - enhancing robustness
- fflas-ffpack-2.3.0, fflas-ffpack-2.3.1 and fflas-ffpack-2.3.2:
  - new code for exact symmetric triangular factorization, triangular inversion, parallel matrix-vector product, etc
  - improved build system
  - enhancing robustness
  - major cleanup
  - increased coverage of the test-suite
- linbox-1.5.0 linbox-1.5.1 and linbox-1.5.2:
  - enhancing robustness
  - reworked polynomial interface
  - improving build system
Integration of these latest releases in SageMath:
- trac #24214
- fixing several long lasting bugs
- speeding-up end-user routines: e.g. the characteristic polynomial of a dense matrix over a modular finite field.
[ISSAC’18] A new recursive algorithm for the symmetric indefinite triangular factorization revealing the rank profile matrix (article submitted and implementation available upstream in fflas-ffpack)
[ISSAC’17] New protocols for the probabilistic interractive certification of the signature, the determinant and the rank profile matrix (extended version submitted to J. of Symbolic Computation).
[submitted paper] New interactive certificates for linear algebra over univariate polynomials
[submitted paper] New protocols secure multiparty matrix multiplication based on Strassen’s algorithm
Prototypical MPI-distributed rational solver.

In collaboration with V. Delecroix (CNRS partner)
- Further exposition of LinBox routines in SageMath.
- Consolidation and improvement of the rational system solver in LinBox
- setup a working jupyterhub VRE using multithreaded matrix multiplication on a 32 cores server for Milestone 8 demo
MPI version of the Chinese remainder based rational solver
DSL for exposing parallel features