Skip to content
/ quadlods Public

Low-discrepancy sequences based on quadratic irrationals

License

LGPL-3.0, GPL-3.0 licenses found

Licenses found

LGPL-3.0
COPYING.LESSER
GPL-3.0
COPYING
Notifications You must be signed in to change notification settings

phma/quadlods

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Overview

Quadlods is a library for making Richtmyer low-discrepancy sequences using quadratic irrationals, which are approximated with multiple-precision rationals. Quadlods also generates Halton low-discrepancy sequences.

Quadlods can generate sequences up to 6542 dimensions (the number of primes less than 65536), but some pairs of primes give particularly bad distributions with high discrepancy. Each prime corresponds to a quadratic irrational (quad for short). The primes are sorted by the maximum continued fraction term of their quads. Use at most 3624 dimensions to avoid the bad pair (13691,13693).

Quadlods scrambles sequences using any of these scrambling methods: third, Thue-Morse, and Gray for Richtmyer, and power, Faure, and tipwitch for Halton. You can also specify none or default.

Compiling

To compile, if you're not developing the program:

  1. Create a subdirectory build/ inside the directory where you untarred the source code.
  2. cd build
  3. cmake ..
  4. make

If you are developing the program:

  1. Create a directory build/quadlods outside the directory where you cloned the source code.
  2. cd build/quadlods
  3. cmake <directory where the source code is>
  4. make

It makes a file called primes.dat, which contains the primes in order of their continued fraction terms, and a file permute.dat, which contains the permutations for tipwitch-scrambling Halton, and installs them in share/quadlods/.

GMP is required both to compile and to run the library. Boost is required to run the program, but not the library. To compile your own code that uses the library, first run "make install" as root. If you are using CMake, put the provided file FindQuadlods.cmake in your project's cmake/Modules directory, along with the GMP and GMPXX modules, and add the following lines to your CMakeLists.txt file:

set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake/Modules/") find_package(Quadlods REQUIRED) set(LIBS ${LIBS} ${Quadlods_LIBRARY} ${GMP_LIBRARY} ${GMPXX_LIBRARIES}) target_link_libraries( ${LIBS}) include_directories(${PROJECT_BINARY_DIR} ${Quadlods_INCLUDE_DIR})

A program called "quadlods" accompanies the library. With it, you can test a set of primes for suitability. There are three tests: scatter, circle, and fill. Scatter and circle test pairs of primes; fill shows how the tuples fill s-dimensional space.

Commands

quadlods sortprimes and quadlods scramblehalton produce the program's data files.

quadlods test and quadlods longtest test the program.

quadlods quadplot plots the quadratic irrationals in a polar graph. r is the square root of the index in A322289; θ is the irrational mod 1, times 2π.

quadlods scatter produces a scatter plot for each pair of quads; a 10-dimensional set produces 45 pages. Recommended number of iterations is 30000, with the plot getting too dark after 100000. You can see the effect of scrambling.

quadlods circle plots the error of finding the area of two quarter-circles with each pair of quads. Again, the number of pages is a triangular number. Good pairs have a variety of forms and the error does not exceed 21.25 and is usually less than 5; bad pairs have a graph like one side of a trumpet, with a large error. Recommended number of iterations is 1048576.

quadlods fill picks s points in s-space at random without replacement and plots the statistics of vectors from each to the nearest generated point. The first graph is deprecated, but I won't remove it until the second is calibrated. The second plots the size of a ball centered at a random point and just touching the nearest low-discrepancy point. The third plots how well the directions from the random points to the nearest low-discrepancy points are distributed on the s-1-sphere. It is normalized so that it should be 1 on average. Recommended number of iterations is 1048576, unless s is high. For high s (like at least 30), the left side of the graphs is unreliable, as most random points are 1/512 or 511/512 in at least one coordinate, and no LD point has yet fallen between the random point and the nearest wall.

quadlods flower draws a flower plot of each component of a sequence. Recommended number of iterations is 30000.

quadlods fourier plots the Fourier transform of each component of a sequence.

quadlods discrepancy computes a lower bound of the discrepancy of a sequence. If two runs on the same sequence give the same number, and no run gives a larger number, it's probably the true discrepancy.

quadlods discplot plots the lower bound of the discrepancy versus the number of points.

quadlods textout outputs a sequence in text.

quadlods interact enters interactive mode, which can be used as an Internet server using xinetd or by programs in any language.

About

Low-discrepancy sequences based on quadratic irrationals

Topics

Resources

License

LGPL-3.0, GPL-3.0 licenses found

Licenses found

LGPL-3.0
COPYING.LESSER
GPL-3.0
COPYING

Stars

Watchers

Forks

Packages

No packages published