Release version 0.10.2 (April 22th, 2026):
    - Bugfixes:
        - Minor fix for upper-bounding solver logging (ipopt)
        - setup.py has been fixed to correctly handle version numbers that contain two digits

Release version 0.10.1 (March 26th, 2026):
    - Improved Babbase Manager performance for large numbers of nodes
    New availability:
        - Added new intrinsic function: Scaled Exponential Linear Unit (SELU)
        - The intrinsic function single_neuron now supports different activation functions
    - Third-party libraries:
        - Fix CMake deprecation warnings related to policy settings and minimum required version
        - Fix CMake warning related to policy CMP0111

Release version 0.10.0 (March 25th, 2026):
    - New features & extensions:
        - New features of GPU Subdomain Lower Bounding: FP32 interval arithmetic, improved GPU/CPU upper bounding, CUDA & memory optimizations, minor fixes
        - Added formatting check to GitLab CI/CD script
    - Thid-party libraries:
        - Upgraded CLP to Release 1.17.10 using CoinUtils Release 2.11.12
    - Bugfixes:
        - Corrected handling for nodes reaching minimum node size
        - Avoid repetitive runs starting from center point in multistart
        - Fixed handling of linear objective within linearization points strategy LINP_KELLEY
	- Minor adjustments of logged output
        - Minor bug fixes

Release version 0.9.1 (July 25th, 2025):
    - Bugfixes:
	    - Fix GitLab CI/CD script to publish documentation automatically

Release version 0.9.0 (July 24th, 2025):
    - New extensions & interfaces: 
	    - New GPU parallel lower bounding solver based on subinterval arithmetic; a CPU-Serial version is also available 
	    - Added an interface to Gurobi, enabling Gurobi as a subsolver in MAiNGO
    - Bugfixes:
	    - Minor bug fixes
    - Third-party libraries:
	    - Upgraded to pybind11 version v3.0

Release version 0.8.4 (May 9th, 2025):
    - Minor bugfixes to the public repository

Release version 0.8.3 (April 22th, 2025):
    - New availability:
        - New intrinsic functions: single_neuron, i.e. custom relaxation and envelope for a single Neuron
        - Updated the optional extension "B&B algorithm with growing datasets"
    - Bugfixes:
        - Minor bug fixes

Release version 0.8.2 (January 23th, 2025):
    - New features:
        - Added example for optimization of multifidelity Gaussian processes using MeLOn
        - Added compliance with PEP 517: python wheels are now installable with Poetry, Pixi, etc.
    - Bugfixes:
        - Minor bug fixes
        - Bug fixes in extension for handling parameter estimation problems
    - Third-party libraries:
        - Upgraded to Nlopt version 2.9.1

Release version 0.8.1 (July 23th, 2024):
    - Bugfixes:
        - Added missing commit in babbase

Release version 0.8.0 (July 23th, 2024):
    - New features & extensions: 
        - New algorithm for specialized B&B algorithm for Two-Stage Stochastic Programming problems, see documentation under Two-Stage Stochastic Programming
        - Extension for handling parameter estimation problems considering large datasets
    - Small bugfixes

Release version 0.7.4 (April 16th, 2024):
    - Bugfixes:
        - Update MathJax to version 3.0 to see latex formulas in the documentation on GitLab Pages

Release version 0.7.3 (April 11th, 2024):
    - Documentation & Testing:
        - Updated MAiNGO documentation
        - Added new unit tests and test cases
    - Bugfixes:
        - Various bug fixes
    - Third-party libraries:
        - Added support for newer CPLEX versions up to 22.1.1
        - Upgraded to Pybind11 version 2.12.0

Release version 0.7.2 (October 31th, 2023):
    - Bugfixes:
        - Fixed minor bug in algorithm for large parameter estimation problems
	- Fixed bug in function of euclidean norm

Release version 0.7.1 (July 26th, 2023):
    - Bugfixes:
        - Added missing commmit in babbase

Release version 0.7.0 (July 26th, 2023):
    - New features & interfaces:
        - MAiNGO can also be used in parallel in python (requires package mpi4py)
        - MAiNGO is now also available in JuMP
    - Bugfixes:
        - In Utility MAiNGOReaderWriter
        - In writing into ALE syntax

Release version 0.6.0 (May 26th, 2023):
    - New features & interfaces:
        - MAiNGO now has a C-API. It mainly expects text input in ALE syntax.
        - New algorithm for large parameter estimation problems, see documentation under Parameter estimation
        - Updates in ALE syntax
    - MAiNGO Algorithm:
        - Now allowing OBBT to only be done with a probability that decrases exponentially with the depth of the given B&B node
        - Minor improvements to rounding heuristic in mixed-integer part that may help find feasible solutions earlier
        - Introduced lazy evaluation for quadratic problems
        - Added intrinsic function mid(x,y,k), where k is a constant and x and y are variables
        - Added intrinsic funciton pinch(x,y,z) = max(x,z)-max(y,z)
    - Bugfixes:
        - Fixed bug that could lead to a (incorrectly calculated) negative final optimality gap
        - Fixed several minor bugs & compiler errrors/warning on different systems
        - Fixed potential segmentation fault im MAiNGO_Reader_Writer
    - Misc:
        - Added tests for MPI-parallelized version
        - Added support for newer CPLEX versions up to 20.1
    - Third-party libraries:
        - Upgraded to Pybind11 v2.10.2
		
		
Release version 0.5.0 (June 14th, 2021):
    - New features & availability:
        - MAiNGO is now also available from PyPI for use via Python
        - New intrinsic functions: fstep & bstep, i.e. unit steps from 0 to 1 (or vice version) at x=0
    - Bugfixes:
        - Fixed bug that caused instant crashes of the MPI parallelized version on some systems (see info on MUMPS below)
        - Minor fixes to ensure compatibility with GCC 11
        - Several fixes in MAiNGO_Reader_Writer utility to avoid compile errors with different versions of GCC and MSVC 2019
    - Misc:
        - Now using GitLab CI/CD for automated testing
        - The documentation of MAiNGO is now hosted via GitLab Pages (see links in Readme.md)
        - Now giving more information on third-party software (local / linear solvers etc.) used
        - Now printing more comprehensive information about the initial point (constraint residuals etc.) when using BAB_verbosity = VERB_ALL
    - Third-party libraries:
        - Upgraded to MUMPS 5.4.0; also renamed all routines called MPI_* to FPI_* to avoid issues with the fake MPI implementation of MUMPS
          when using MAiNGO with actual MPI parallelization


Release version 0.4.0 (March 4th, 2021):
    - New features:
        - MAiNGO now has a Python API. It consists of Python bindings for the C++ API and thus works very similarly to the latter
    - Examples & documentation:
        - Added more information on output (screen & files) and algorithm of MAiNGO
        - Added information on Python interface
        - Added an example for the Python API to examples/01_BasicExample
    - Misc:
        - Various bugfixes (e.g., avoiding potential crashes of the B&B or in the parallel version etc.)
        - Improved error reporting
        - Renamed a few options as well as methods of the MAiNGO class to be more descriptive
        - Fixed random seeds for CPLEX and CLP
        - Improved diagnostic output for problems without objective or with constant objective
    - Third-party libraries:
        - Included pybind11 for the Python API
        - New version of babbase containing minor bugfixes, and now also allowing binary variables without explicit bounds
        - New version of mcpp containing bugfixes in relaxations and constraint propagation


Release version 0.3.0 (June 12th, 2020):
	- New features:
		- MAiNGO now uses the toolkit MeLOn - Machine Learning Models for Optimization
			- MeLOn contains tools for modeling and training different machine learning models such as artificial neural networks or Gaussian processes
			- The models from MeLOn can now be used when writing problems for the C++-API of MAiNGO
			- These models use some custom relaxations available in MAiNGO through MC++ for faster convergence
			- In order to use MeLOn within MAiNGO, the Cmake flag MAiNGO_build_melon needs to be set to true (default)
			- Note that MeLOn is not compatible with current Intel Compilers due to missing C++17 features
	- MAiNGO algorithm:
		- The parser now preserves the order of variable declaration and does not eliminate variables that do not occur in the problem
		- Added new intrinsic functions (for more details, see doc/implementedFunctions/Implemented_functions.pdf):
			- Common acquisition functions for Bayesian optimization:
				- Lower confidence bound
				- Expected improvement
				- Probability of improvement
	- Examples & documentation:
		- Examples were added for the use of the models from MeLOn:
			- An example for problems with artificial neural networks embedded
			- An example for problems with Gaussian processes embedded (e.g., minimizing the prediction or variance of a Gaussian process)
			- An example for using MAiNGO within Bayesian optimization (i.e., maximizing/minimizing typical acquisition functions for Bayesian optimization that use Gaussian processes)
		- The documentation for building MAiNGO was improved
		- A section on the output written by MAiNGO was added
	- Third-party libraries:
		- Included MeLOn (see above)
		- Upgraded to new MUMPS version 5.3.1.


Release version 0.2.1 (February 17th, 2020):
	- Examples & Documentation:
		- Updates in readmes and manual
		- Moved example problems to folder 'examples' & added readmes for example
		- Bi-objective optimization:
			- Added utility for plotting Pareto fronts
			- Added examples for epsilon-constraint method for bi-objective problems
		- Improved output of MAiNGO (e.g., for Branch-and-Bound; missing input files; ...) and CMake
	- MAiNGO algorithm:
		- Added new intrinsic functions (for more details, see doc/implementedFunctions/Implemented_functions.pdf):
			- covariance functions for Gaussian processes
			- function 'regnormal'
			- Gaussian PDF functions
		- Fixes in algorithms
		- Fixes in parser
		- Fixes in computation of relaxations (through update of MC++).
	- MAiNGO_Reader_Writer utility:
		- Now reading dict.txt file generated by GAMS convert to retain original variable and equation names (or similar).


Release version 0.2.0 (November 8th, 2019):
    - Initial release