The PDF documents at our site are copyrighted but not subject to the liberal terms of the Boost License . You may download them for your personal use and you may set hyperlinks to them; however, the documents may not be posted online by anyone other than us. These are living documents that are updated over time. Unauthorized copies posted to the Internet run the risk of not being updated by the poster when we update ours.

 April 25, 2015. Removed documents that are not accessed frequently or need significant rewriting. CommandLineParsing.pdf ("Command Line Parsing") CollinearityTest.pdf ("A Collinearity Test Independent of Point Order") PdeNumericalMethods.pdf ("Numerical Methods for Partial Differential Equations") OdeNumericalMethods.pdf ("Numerical Methods for Ordinary Differential Equations") DistancePointToQuadratic.pdf ("Distance from Point to a General Quadratic Curve or a General Quadric Surface") April 24, 2015. Expanded the document to include more NURBS representations. In particular, there is a degree-3 representation for a full circle that uses an internal repeated knot to splice together two half-circles. There is a tensor product NURBS surface, degree-3 in each parameter, that uses the same idea of an internal repeated knot to splice together two hemispheres. April 23, 2015. Major rewrite of the document on converting between coordinate systems. The original version of this document was entitled Conversion of Left-Handed Coordinates to Right-Handed Coordinates and was written to handle the conversion of LightWave coordinate systems (left-handed) to Wild Magic coordinate systems (right-handed). The process was specific to LightWave's choice of representing rotations using Euler angles, and the discussion included how to deal with cameras, lights, and transformation hierarchies. The new version is the general process of converting between any two coordinate systems. An implementation is provided that automates the process. The old file, LeftHandedToRightHanded.pdf, was deleted. April 20, 2015. Added significant improvements to the document for the test-intersection and find-intersection queries for ellipses and to the document for computing the area of intersection of ellipses. The source code implementation for the test-intersection and find-intersection queries were updated to use the details described in the document. Implemented the query for area of intersection of ellipses. April 4, 2015. Rewrote the document for computing the roots of low-degree polynomials. The new document goes into great detail about the classification of roots (real or non-real, multiplicities) and how to use exact rational arithmetic to correctly classify the roots in a program. This leads to more robust root finding using the closed-form expressions for polynomials of degrees 2, 3, and 4. The motivation for the revisions was based on trying to compute intersections of ellipses, and the nonrobust root finder for quartic polynomials created many problems numerically. April 2, 2015. Fixed the algorithm description for conversion from a BSRational object to a floating-point type. When the round-up step causes a carry-out, so to speak, from the trailing significand, a block of code was executed to set w to 1 and adjust p-q. This was incorrect and, in fact, not necessary because w is not used as the trailing significand in the conversion. March 29, 2015. Added some definitions for variables used in the equations in the section {\em Separation Tests Involving Other Directions}. Converted the verbatim pseudocode to use lstlisting format. March 10, 2015. Fixed typographical errors and added some clarification. January 26, 2015. Fixed an error in the description for region 2 and added some clarification about the sign tests for the partial derivatives of R(s,t). January 5, 2015. Added comments about fitting lines y = a*x + b and planes z = a*x + b*y + c, mentioning that the numerical problem for solving the linear equations in the coefficients is ill conditioned unless you compute the averages of the data points first and subtract those averages from the data points before solving the linear system. January 4, 2015. A new document describing the test-intersection query for boxes and cones. December 28, 2014. Modified the document on line-cone intersection to give greater detail. The source code was updated accordingly. December 12, 2014. The document was rewritten to add more details about the algorithm. The source code was modified accordingly. December 7, 2014. A new document describing a variation of an iterative eigensolver for symmetric $3 \times 3$ matrices. A source code implementation and a sample application are provided. The related document Eigensystems for 3x3 Symmetric Matrices (Revisited) (EigenSymmetric3x3.pdf) has been deleted and that link redirected to the new file. The document Eigensystem Solvers for Symmetric Matrices has also been deleted. The GTEngine code now uses only the iterative solvers for symmetric matrices; we also have an iterative implementation for the singular value decomposition (SVD). November 27, 2014. The pseudocode for computing the fitted cylinder subtracted the input point average for numerically stable computations. The returned center needed the average added to it. November 25, 2014. Overhauled the arbitrary precision library to improve the performance and to improve readability. The unsigned integer arithmetic was factored out of BSNumber into two classes, one for arbitrary precision with storage of type std::vector and one for user-selected fixed precision with storage of type std::array. Both classes share code for the arithmetic logic unit. Many computations are now performed in-place to avoid expensive allocation, deallocation, and memory copies. A new PDF is posted that greatly expands on the library compared to the discussion in the GPGPU Programming for Games and Science book. The document serves as a discussion about the design of the library and a reference for how to use it. Examples are provided for using BSPrecision to determine the template parameter of UIntegerFP that represents the maximum number of bits required to compute the exact results for a sequence of expressions. November 6, 2014. Implemented a robust algorithm for computing the distance between line segments in any dimension. Revised the PDF for computing distance between segments in 3D to describe the new algorithm. A GPU implementation is available in the sample application. October 20, 2014. The volume equation for a dodecahedron was in error. August 15, 2014. Fixed a typographical error in Equation (14). Replaced the \LaTeXe verbatim commands with lstlisting commands for more readable pseudocode.