• 2016-06-23 v2.4.0
• Introduced nomeclature 'mode'. Marked all functions that will be renamed / removed in v3.0.0 as deprecated.
• Added an experimental python wrapper when 'OTHER += -DXERUS_EXPERIMENTAL_PYTHON_WRAPPER' is defined in the config file.
• Added TT::degrees_of_freedom function to determine the dimension of the current tangent space.
• ! Corrected a typo in the class names of the measurement sets.
• Added constants xerus::VERSION_MAJOR, VERSION_MINOR, VERSION_REVISION and VERSION_COMMIT.
• Improved compilation time and more forgiving default settings for warnings.

• Added save_to_file and load_from_file functions to store and restore Tensors and TensorNetworks in/from files.
• Added SVD based calculate_pseudo_inverse function to Tensor.
• Added implementation for perform_trace() for sparse Tensors.
• Fixed the pipe operator for xerus::misc::generic_exception to not destroy the type information.
• Bug fix in the handling of fixed indices in TensorNetworks.
• Several static member function now warn if their return type is not used.
• Initial support for compilation with the intel ICC.

• Added support for 32bit systems.

• Added Sparse solve and sparse QR factorization.
• ! Xerus now depends on the SuiteSparseQR (-lspqr) library.
• TTNetwork sum now supports sparse components.
• TTNetwork entrywise_product now supports sparse components.
• ! entrywise_product (for TTNetworks) now a free function instead of static member.
• TTNetwork entrywise_square now supports sparse components.
• Bugfix in Dense/Sparse conversion factor handling.

• Basic support for fixed indices in TensorNetworks.
• Readded TTNetwork constructor from DimensionTuple.
• Bugfixes for TTNetwork dyadic_product.

• ! fix_slate() now works correctly for TTTensors and (correctly) throws an error for TTOperators.
• Several bug fixes in the handling of TTNetworks of degree zero.
• Implemented the Tensor CQ function, which is now used where appropriate.
• Added several unit tests to ensure consitent results for dense/sparse Tensors, TensorNetwork and TTNetworks.

• ! Merged FullTensor and SparseTensor into a single Tensor class.
• Implemented several FullTensor functions also for sparse representations (modify(_diag)_elements, fix_slate, resize_dimension,...)
• ! IndexedTensor*s now considered internal.
• ! Removed all support for stored IndexedTensor*s.
• ! Reworked the IndexedTensor* index assignment.
• ! Changed resize_dimension to allow slate insertion at the first position.
• Added TTTensor::random with callback function to manipulate the singular values of all matricisations.
• Rewrote the ALS algorithm for better readability.
• Added fully functional (multi-site) DMRG and alternating steepest descent algorithms.
• Support for low level factorisation calls for Tensor class.
• Several bug fixes, including SVD factor handling, SparseTensor summation, Tensor resize_dimension, TN evaluation,....
• Added several test cases.

• Added the ADF algorithm for rank one measurements instead of point evaluations.
• Parallelized the ADF algorithm using openmp (note: at this point not compatible with the replacement allocator!)
• TT::random no longer uses size_t to measure the size of the tensor and no longer exceeds maximal ranks.
• Fixed a bug in the ALS that caused the algorithm to always stop after one sweep if called without operator A.

• The custom allocator now enforces memory alignment even if the underlying malloc does not.
• Cannonicalized TT tensors can no longer exceed the maximal TT rank.

• Fixed a bug with the recannonicalization of products of TTOperators.
• Many improvements for the ADF algorithm for tensor completion.
• Added TTTensor::ones and TTOperator::ones to create TTTensors with all entries equal to 1.

• Added an implementation of the ADF algorithm.
• Added a custom allocator to better handle the small allocations of Indices etc. (requires to set the corresponding option in the config.mk).
• Reduced overhead (time not spent in blas) of TensorNetwork contractions.
• Tensor decomposition now allow arbitrary index orders in the target as well as the source.
• Added TN::measure(...) to perform a batch of evaluations of the network.
• Added TN::fix_slate(...) to fix a single index to a specific value.
• Added TN::reduce_representation() to trace out any full-rank edge.
• Added TN::transfer_core(...) and TN::round_edge(...) analogously to TT rounding / core movement operations.
• ! renamed class::construct_random to class::random and TTOperator::construct_identity to TTOperator::identity.
• Added several named constructors:
• Tensor::ones(...) to construct tensors filled with all entries equal to 1.
• Tensor::identity(...) to construct a (sparse) tensor equal to the TTOperator::identity.
• Tensor::dirac(...) to construct a (sparse) tensor with a single 1 at a given position and 0 everywhere else.
• Tensor::kronecker(...) to construct a tensor that is 1 when all indices are equal and 0 otherwise.
• ! misc::approx_equal now uses a relative difference (for consistency with the tensor variant xerus::approx_equal).
• Several bugfixes for SparseTensors.
• Several minor bugfixes (TN::draw no longer uses temporary files, span-zero indices in assignment, ...).

meant for use by developers only!

• Added several retractions to allow Riemannian algorithms on TT manifolds.
• Added the basic (Riemannian) steepest descent algorithm.
• Added the (so far untested) CG algorithm for TT tensors.
• ! New PerformanceData class to log stepcount, time and residual of algorithm runs.
• New algorithm to determine the largest entry of a TT tensor.
• Fully rewrote the contraction heuristics, now including a local search for more optimal contraction orders.
• Several bugfixes for order-zero tensors, TT tensors of more than full rank and other rare cases.

• entrywise_product for FullTensor and TTNetwork.
• TensorNetworks now without extraneous factor.

• Rank revealing QR decomposition in TTNetwork:move_core.
• ALS performance improvement by skipping full-rank nodes.
• Added TensorNetwork::draw to get a graphical representation of a given tensor network.
• Added algorithm DecompositionALS to approximate a FullTensor by a TTTensor.
• New Quick-Start guide.
• Several bugfixes, more consistent operator availability and much more documentation.

• Complete documentation for Tensor, FullTensor and SparseTensor
• Added implementation of the projection ALS
• Fixed performance bug in logging system
• ALS now uses energy functional for convergence testing
• Several minor fixes

• ! New representation for the TT tensors: all component tensors are now of degree 3 / 4 (operators).
• TT::get_component and ::set_component functions that handle the meta data.
• More strict ISO c++ conformity: no names starting with '__'.
• More elaborate performance profiling.