Skip to content

Latest commit

 

History

History
139 lines (101 loc) · 5.96 KB

README.md

File metadata and controls

139 lines (101 loc) · 5.96 KB

Polytopes, lattices and quadratic forms programs

This is the set of functionality for dealing with polytopes, quadratic forms and lattices.

The general approach is to use polytope and polyhedral structures and groups are used to make everything faster as a general rule.

The goal is to get extreme speed in order to solve record problems.

Access to the source code

Since this repository uses submodules, the cloning command is

$ git clone https://github.com/MathieuDutSik/polyhedral_common --recursive

In order to update the submodule the command is

$ git submodule update --remote

Compilation

The compilation of the software is relatively complex. However in script_docker/Dockerfile a dockerfile is given that should make everything clear. It should also allow any user to install the system on their computer and use it fairly easily.

General organization of the code

The program are in several independent subdirectory. The software of each directory can be compiled independently of the others:

  • src_group: for computting the groups of a polytope.
  • src_poly: for polyhedral computations.
  • src_dualdesc: for computing dual description on serial computers.
  • src_copos: for copositivity / strict copositivity functionalities.
  • src_short: for short vector related computations.
  • src_latt: for lattice related computations (canonicalization of positive definite matrices and shortest vector mostly)
  • src_ctype: for computing C-types. This was done for computing all the C-types in dimension 6.
  • src_perfect: for perfect form related computations.
  • src_indefinite: for indefinite form reduction.
  • src_lorentzian: for using the Vinberg/Edgewalk algorithm of hyperbolic forms.

Compilation options related to debug

There are several environment variables that can be used during runtime.

  • DEBUG for making some print statements and making some checks. Those checks have to be fast.
  • KEY_VALUE for printing some KEY(....) VALUE=(....) that can be used for postprocessing of the options and heuristic optimization.
  • CHECK for making checks that can be expensive to do.
  • TIMINGS for printing some runtime information.
  • SANITY_CHECK for doing some checks and stopping if incoherence are detected.

The options TIMINGS and DEBUG enable all the timings and debugging statements. For a more granular debugging, stuff like DEBUG_LINEAR_PROGRAM can be used. See the top of the header files.

A printout to std::cerr should occur if an error has been identified and the program will terminate with a call to TerminalException. Other print statement should be encapsulated in std::ostream & os that should be passed by reference from the initial call. So typically for serial output we pass the std::cerr while for the parallel runs we pass a stream to the output of that process. That way we avoid mixing between different sources.

Arithmetics

Many of the programs of this repository allow to take different arithmetic. This is both from a functional aspect and a speed aspect. There is no systematic way this is done but the following rules are applied. The types are always template parameter.

Functional types:

  • Qsqrt2, Qsqrt3 and Qsqrt5. Those types are used for the fields Q(sqrt(2)) and similar. For each field we need a type. So, if you need say Q(sqrt(6)), you need to modify the code.
  • The function RealAlgebraic=FileDesc is for working with real algebraic numbers with the description in FileDesc. That file should contain the continous fraction approximant up to some chosen precision. If insufficient, a clean error of failure will be reported.

For both those types, the coefficient are written in entries like 1, 1+x, x/4, -3-3x/4, *3+(3*4)x^3 or such. No space in the entry.

Speed types:

  • The mpq_class is from gmp and is the standard rational type being used.
  • The mpz_class is from gmp and is the standard integer type being used.
  • The boost::multiprecision::cpp_rational is the pure boost based rational type. Not as fast as gmp, but purely heaer based.
  • The boost::multiprecision::cpp_int same as above but for the integers.
  • The boost::multiprecision::mpq_rational is boost based encapsulation of gmp library.
  • The boost::multiprecision::mpq_int same as above but for the integers.
  • The Rational For the rational implementation from an integer type.
  • The SafeInt64 for integer computation in int64. If the computation go over the limit, then an exception is raised and typically the computation stops. But no wrong result is reported.
  • The Rational the same as above but for rational fields.

The choice is usually for mpq_class which has the least issues.

Tests

The directory Exmpl_Bench contains a bunch of test and development. They are typically work in progress and not necessarily in a state of being finished.

The directory CI_tests contains some tests that are run in CI on GitHub. Their runtime should be short, from 5 minutes to 1 hour. Together they should cover as much as possible of the functionality of the code. If the test is working like normally, then it should be scheduled in the cron to run once per month. We do not want to overflow the credit that we have.

Usage

The primary use of the programs is via the CLI. But this can be relatively impractical.

Therefore, we plan to have interfacing with common Computer Algebra Systems:

  • GAP (Group Algebra Programming) which is an independent system.
  • Oscar which is based on Julia.
  • Sage which is based on Python.

Dependencies

Following dependencies are needed for compiling the code: