Parallelization of numerous algorithms using OpenMP

• Version 7.0 of the IMSL C Library enables customers to better take advantage of multi-core and many-core hardware for improved performance. Numerous algorithms leverage OpenMP directives on supported environments to distribute calculations across available resources.
• The following figures show performance comparisons for several algorithms across multiple cores. For each of these benchmark comparisons, a dual quad core Xeon E5420, 2.5GHz with Windows Server 2003 R2 was used. To see a more comprehensive set of benchmarks, download the “Parallel Performance of the IMSL C Library” white paper.

Figure 1: Random number generation algorithms showing performance, relative to one thread

Figure 2: Numerical optimization algorithms showing performance, relative to one thread

New function that solves the generalized Feynman-Kac PDE and Black-Scholes problems

• Solves a generalized version of the Feynman-Kac partial differential equation that can be used in many financial modeling applications, including the Black-Scholes models with European or American style exercise opportunities on Calls or Puts. In the case of the Black-Scholes model these functions include many of The Greeks.
• Two new white papers are available that provide more detail on this algorithm:
• "Integrating Feynman-Kac Equations Using Hermite Quintic Finite Elements" is a technical white paper that describes the method for solving the Feynman-Kac PDE
• "Solving Constrained Differential-Algebraic Systems Using Projections" is another technical white paper that describes specific software additions to the Differential-Algebraic Equation (DAE) solver code

New data mining functions including a Genetic Algorithm for optimization and Naïve Bayes for classification problems and text mining

• The IMSL Library Genetic Algorithm implementation supports the basic algorithm originally introduced in the 1970s with the most popular variations. This capability is achieved by supporting variations such as user defined population size and selection methods, random or user defined initial populations, any combination of four different data types: nominal, binary, integer and real, and user supplied fitness functions with or without additional function parameters. These many variations offer developers a level of flexibility in the IMSL Library Genetic Algorithm functions that is unmatched in alternative solutions.
• Naïve Bayes is a simple algorithm that is very fast. A Naïve Bayes classifier can be trained to classify patterns involving thousands of attributes and applied to thousands of patterns. As a result, Naïve Bayes is a preferred algorithm for text mining and other large classification problems.

Many other new functions, including:

• Kochanek-Bartels Cubic Splines
• Non-central chi-square, Non-central student’s T PDFs

Enhancements to many existing algorithms, including:

• Improved algorithm for finding zeros of a function
• Faster normal random number generation
• Neural network classification capability
• Multiple options for selecting Auto_ARIMA models