Improvements and Updates

What's New in IMSL

Get Email Updates Get Latest Version

What's New

With proven, tested algorithms and functions from Perforce IMSL, your team can call, embed and test an algorithm in a fraction of the time, saving your team a significant amount of development time and money.

Back to top

IMSL For Fortran (FNL) 2026.1

We're pleased to announce the immediate availability of IMSL FNL 2026.1, which includes important fixes, enhancements, and updates designed to improve reliability and performance across the library. Keep reading for a summary of what's new, or visit the release notes directly via the link below.

FNL 2026.1 Change Log  FNL 2026.1 Supported Platforms

Improved Optimization Routines

We've addressed several issues in the TRAN transportation optimization routine, including:

  • Corrected the sign of the dual solution returned through the optional DUAL argument.
  • Improved the implementation of the revised simplex method for transportation problems where total warehouse capacity is less than total store demand.

These updates help ensure more accurate optimization results across a wider range of real-world scenarios.

Strengthened Linear System Error Checking

For the `LSLXD` and `L2LXD` routines in the Linear Systems chapter, we've added error checks that detect insufficient workspace allocation, helping catch configuration problems earlier and preventing unpredictable behavior downstream.

Enhanced Statistical Accuracy

For the UVSTA routine in the Basic Statistics chapter, we've corrected the default handling of the nvar parameter when frequency (ifrq) and/or weight (iwt) arguments are provided. This improvement delivers more reliable statistical calculations and expected behavior in weighted and frequency-based analyses.

Multidimensional Scaling Improvements

We've resolved several accuracy and stability issues in the Multidimensional Scaling routines:

  • MSIDV — Corrected internal computational errors affecting the numerical accuracy of the individual-differences multidimensional scaling algorithm, fixed incorrect variable indexing in error handling, corrected weighting factors in gradient and Hessian computations, and improved conditional printing behavior.
  • MSINI — Corrected an internal variable indexing error in regression-based initial estimate calculations.
  • MSSTN — Fixed the reciprocal conversion (`ICNVT`=2) for similarity-to-dissimilarity transformation by correcting array indexing and initialization.

Together, these fixes improve the reliability of individual-differences scaling analyses.

Deprecation Notice for Sparse Linear System Routines

The following routines have been deprecated and will be removed in a future release:

LFTXG, LFSXG, LSLXG and their complex-precision equivalents (LFTZG, LFSZG, LSLZG)

We recommend transitioning to the linear_operators module using the .ix. operator with the d_hbc_sparse and s_hbc_sparse derived types (or z_hbc_sparse and c_hbc_sparse for complex operations). Please refer to the Sparse Matrix Computations documentation for detailed guidance.
 

Platform Updates

The FNL 2026.1 support matrix adds RHEL 10 support (with GCC 14.2.1 and Intel oneAPI 2024/2025/2026), Intel oneAPI 2026 (IFX) across all Linux and Windows platforms, and drops RHEL 8, Windows 10, and Intel oneAPI 2023 across all platforms. The net effect is modernizing the OS baseline and shifting the compiler window up to the latest 3 releases.

Back to top

Highlight improvements in IMSL C 2026.1.0 include support for popular machine learning platforms, including oneAPI 2025.3, Visual Studio 2026, and RHEL 10. 

In addition to platform work for this release, it also includes:


Probability Distribution Functions and Inverses

  • weibull_pdf, weibull_cdf, weibull_inverse_cdf Evaluates the pdf, cdf, and inverse cdf for the Weibull distribution. The Weibull distribution is widely used in reliability engineering, materials science, weather modeling, quality control, and finance.

Data Mining

  • prefix_span Performs sequential pattern mining using the PrefixSpan algorithm. Sequential Patterns are frequently occurring sequences of items. Such items could be products, genes, behaviors, symptoms of disease, virtually any observable, discrete event. Applications for SPM include problems in medicine, bio-informatics, economics, psychiatry, retail and e-commerce, and many others. The PrefixSpan algorithm is a depth-first search that "discovers" or grows sequential patterns by way of projection and recursion.

The full list of the latest supported platforms is available online.

Change Log 

Back to top

IMSL For Fortran 2025.1

IMSL (R) Fortran Numerical Library provides the essential building blocks needed to develop analytic applications for your organization. Recent updates to IMSL (R) Fortran 2025.1 provide new features and bug fixes, including added nonsingularity check for the U factor in array FACT in the LAPACK case. 

IMSL (R) Fortran 2025.1 also includes support for the GNU Fortran compiler on Red Hat Enterprise Linux 9 and the Intel oneAPI 2025 compiler.  

The full list of the latest supported platforms is available online.

Back to top
Back to top

IMSL For Fortran (FNL) 2024.1

The latest release of FNL includes several platform updates including RHEL 9, Intel oneAPI 2024.0 and support for the new oneAPI IFX compiler on several platforms. 

The full list of the latest supported platforms is available online.

Back to top
Back to top

IMSL C 2025.1

IMSL C Numerical Library provides the essential building blocks needed to develop analytic applications for your organization. Recent updates to IMSL C 2025.1 provide new features and bug fixes, including three functions for negative binomial distribution. The negative binomial distribution provides a useful model for counts of discrete random events, such as incidences of disease in certain populations, occurrences of tornados across the country, and many others.  

In addition, there is now a graph-based variant of the simplex method normally used to solve the transportation problem. By finding an initial solution relatively close to the optimum, this method often results in fewer steps being required to get the optimal solution. Internal experimentation has shown this method to be often significantly faster than the two other methods available, the revised simplex and the interior-point method. 

Additional features include: 

  • Support for Intel oneAPI 2025
  • Performance improvements and a new optional argument to control tolerance in internal rate of return
  • Functions to calculate the determinant of real or complex square matrices 

For more information, see the release notes. 

Release Notes 

Back to top
Back to top

IMSL C 2024.1

New Features

  • Developers are now able to generate, store, and reuse gradient boosted models in IMSL.  This allows the developer to work in two phases to first generate a model and then use that model in a production environment without having to train the model each time.
  • IMSL now allows you to compute eigenvalues and eigenvectors of symmetric and Hermitian band matrices in band symmetric storage format.

Bug Fixes

  • This release includes several bug fixes and improvements in memory use and error handling.  Updating to this version will help you stay up to date with underlying platforms and provide the most stable environment for your analytics applications.
  • The full list of fixes is available in the readme.
Back to top