version.log

v1.0.0
1. [Important] This is the first stable release of the project.
2. [Change] Merge Contract and Contracts into Contract, and Contract_ and Contracts_ ito Contract_.
3. [Change] Merge relabel and relabels into relabel, and relabel_ and relabels_ into relabel_.
4. [New] Add an optional argument min_blockdim to svd_truncate to define a minimum dimension for each block.
5. [New] Add Eig/Eigh functions for Block UniTensor.
6. [New] Add Lancos-like algoirthm, Lanczos_Exp, to approximate exponential operator acting on a state.
7. [Change] Migrate cuTENSOR APIs to the version 2.
8. [Change] reshape_ and permute_ to return the object itself instead of None.
9. [Change] Remove the magma dependency.
10. [Enhance] Optimize the contraction order finding algorithm.

v0.9.7
1. [New] Add the identity/eye functions for UniTensor to generate an identity UniTensor.

v0.9.6
1. [Fix] Resolve an issue where Trace did not work for is_diag=True.
2. [New] Add support for the UniTensor type for linear algebra function Eig, Eigh and InvM.
3. [Fix] Fix incoorected behavior in uniform random and isub.
4. [New] Add beartype checking for ovld warpper.
5. [Fix] Fix CICD tools.

v0.9.5
Same content as 094, fix CD to conda-forge

v0.9.4
This has the same content as v0.9.3

v0.9.3
1. improvements of GPUs, integrated with cutensor/cuquantum.
2. extra Svd methods for both CPU and GPU.
3. Other improvements of user API including conversion between symmetric and non-sym UniTensors.

v0.9.2
1. [Important] [Change] Remove all deprecated APIs and old SparseUniTensor data structure
2. [Fix] Bugs when batch_matmul when no MKL
3. [Update] Update examples to match new APIs
4. [New] add labels options when creating UniTensor from Tensor.
5. [New] change MKL to mkl_rt instead of fixed interface ilp64/lp64

v0.9.1
1. [New] Add additional argument share_mem for Tensor.numpy() python API.
2. [Fix] UniTensor.at() python API not properly wrapped.
3. [Fix] Bug in testing for BlockUniTensor.
4. [Fix] Bug in UniTensor print info (duplicate name, is_diag=true BlockUniTensor dimension display)
5. [Change] Svd now using gesdd instead of gesvd.
6. [New] Add linalg.Gesvd function, along with Gesvd_truncate.
7. [Fix] Strict casting rule cause compiling fail when compile with icpc
8. [New] Add additional argument for Network.PutUniTensor to match the label.
9. [Fix] Network TOUT string lbl bug
10. [Fix] #156 storage python wrapper cause not return.
11. [Add] linalg.Gemm/Gemm_()
12. [Add] UniTensor.normalize()/normalize_()

v0.7.7
1. [Enhance][WARNING] rowrank option now has default value when converting from Tensor. Which is half number of the bonds. Notice that the order of argument are changed between (rowrank) and (is_diag)!
2. [Fix] Svd will have issue associate to changing of rowrank/is_diag order.
3. [Enhance] Internal Syntax format change to clang format.
4. [Change] USE_OMP option gives openmp access only for in-house implementation. Any linalg funciton calling MKL will be parallel.

v0.7.6
1. [Enhance] Adding alias BD_IN=BD_KET, BD_BRA=BD_OUT, BD_NONE=BD_REG.
2. [New] Add Contracts for multiple UniTensors contraction.
3. [Fix] cytnx.__cpp_lib__ for some version of cmake and conda install, libpath is lib64 instead of lib.
4. [Optimize] SparseUniTensor contiguous (moving elements)
5. [Optimize] cytnx_error_* will now evaluate the clause first, and then instance the following strings.
6. [Enhance] Add Global bool variable User_debug, which when set to false some checking will be skipped, which increasing the execution speed
7. [Enhance] Add Network.getOptimalOrder()

v0.7.5
1. [Fix] ICPC cannot compile issue
2. [Fix] openblas wrapper of zscal has wrong format, cscal,sscal not wrapped (using mkl is not affected)
3. [Enhance] auto_install.py
4. [Enhance] add vec_cast utility.
5. [Fix] Svd_truncate with err does not properly truncate the values.
6. [Fix] MatVec dgemv reversed argument.
7. [New] Add Histogram2d class in stat tools
8. [Enhance] Add SparseUniTensor.Save / .Load
9. [Enhance] Add vec_fromfile / vec_tofile in utility.
10. [Enhance] Adding omp parallel for SparseUniTensor moving elements, and L1-optimized.
11. [New] Add Storage.vector<>() for converting Storage to std::vector.

v0.7.4
1. [Enhance] Lanczos_ER Lanczos_Gnd not convergence with maxiter will now gives warning instead of error.
2. [Enhance] Arithmetic of UniTensor(&)constant now preserve the label of input UniTensor.
3. [New][experiment] Add MPS class with two variant: iMPS, RegularMPS.
4. [New][experiment] Add MPO class.
5. [Enhance] Add UniTensor.relabel
6. [Enhance] Add Network.FromString
7. [New][experiment] DMRG API
8. [New][experiment] Add MPS Save/Load, and can now have different phys_dim for each site.
9. [Fix] SparseUniTensor.permute does not properly update contiguous status when rowrank argument is given.
10. [Enhance] get_block(_)/put_block(_) by qnums now have a new argument "force" to get blocks from non-braket_form UniTensor.
11. [New] Add SparseUniTensor contract
12. [New] Add SparseUniTensor linalg::Svd support.
13. [Enhance] SparseUniTensor print info, add "contiguous" status.
14. [Enhance] Add print_info for Symmetry descriptor
15. [Enhance] Add UniTensor.syms()
16. [Fix] Tensor.set when one of accessor is Singl will cause error.
17. [Enhance] SparseUniTensor diag x diag, diag x dense are finished.
18. [Fix] SparseUniTensor when diag permutation issue.
19. [Fix] Sort does not return out Tensor.
20. [Fix] Tproxy.item() does not get correct element.
21. [Fix] Bug for Svd on SparseUniTensor vT is being set by U
22. [New][experiment] Svd_truncate for SparseUniTensor
23. [New] add Bond.redirect(), Bond.retype()
24. [Fix] SparseUniTensor.permute() does not properly update braket_form
25. [Fix] SparseUniTensor.set_rowrank should track _inner_rowrank not _rowrank bug.
26. [Enhance] Add UniTensor.change_label() <- [Removed!!] use relabel(s)()
27. [Fix] Svd_truncate when one of the block has only dim=1 should fill-in the dangling dimension.
28. [New][experiment] iTEBD with U1 symmetry example for Heisenberg chain
29. [Change] v0.7.4 [26.] replace change_label() with relabel. Now only have set_label(s) and relabel(s) with *_label() have by_label option.
30. [Enhance] Add Accessor option Qns, qns()
31. [Change] Trace now by default trace axis =0 and axis=1 if no argument specify.
32. [Fix] Compare of two Bonds will now also check qnums.
33. [New][experiment] SparseUniTensor.Trace() now support rank-2 symmetric UniTensor -> scalar
34. [New][experiment] Contract of SparseUniTensor with two SUT with same labels -> scalar is now avaliable
35. [Fix] DMRG initialize does not properly normalize the init state.
36. [New] Scalar.conj(), Scalar.real(), Scalar.imag(), Scalar.maxval(dtype), Scalar.minval(dtype)
37. [Enhance] Lanczos internal now written with single general function.
38. [Enhance] Storage.append() now accept Scalar
39. [Enhance][Fix] Fix inplace Arithmetic between Tensor +=(-=,*=,/=) Tensor with both non-contiguous leads to inconsistent memory alignment.
40. [Enhance] from 39. add iAdd(), iDiv(), iMul(), iSub(), this can be called by user but is not recommended.
41. [Enhance] Modify DMRG kernel for generic UniTensor as state.
42. [New][experiment] Add Lanczos_Gnd_Ut() which accept Tin as UniTensor
43. [New][experiment] LinOp now add an matvec option for UniTensor => UniTensor, which can be used together with Lanczos_Gnd_Ut
44. [Change] Remove LinOp with custom function support, inheritance is forced.
45. [Enhance] add Tensor.at() without template.
46. [Change][Enhance] Remove UniTensor.get_elem/set_elem, unify them with at().
47. [Fix] Trace for SparseUniTensor with is_diag=True.
48. [New][experiment] MPS.Norm()
49. [Fix] Lanczos_Gnd_Ut when input dimension is only 2 now check if the beta=0.
50. [New] Add DMRG U1 example.
51. [Change] Behavior change for Svd_truncate. SparseUniTensor the keepdim can exceed the current dimension of UniTensor, in such case it is equivalent to Svd.
52. [New] Add UniTensor.Norm()
53. [New][experiment] add MPS.Init_Msector(), which initialize the state with specify total magnetization.
54. [Enhance] Add additional feature Svd_truncate with truncation_err (err) and return_err option for Ten
55. [Enhance] Add additional feature Svd_truncate with truncation_err (err) and return_err option for DUTen
56. [Enhance] Add python dmrg example for using tn_algo

v0.7.3
1. [Fix] bug for Get slice does not reduce when dim=1.
2. [Enhance] checking the memory alloc failing for EL.
3. [Change] remove Tensor init assignment op from initializer_list, for conflict with UniTensor init.
4. [Enhance] print information for Symmetric UniTensor.
5. [Enhance] linalg::ExpM/ExpH support for symmetric UniTensor.
6. [Enhance] add UniTensor.get_blocks_qnums() for corresponding qnums for current blocks.
7. [Enhance][Safety] add UniTensor.get_blocks_(silent=false) with "silent" option by default pop-up a warning when UniTensor is non-contiguous.
8. [Enhance] add operator* and operator*= for combineBond.
9. [Enhance] add support for Symmetric UniTensor with is_diag=true.
10. [Fix] remove the dtype & device option for arange(Nelem). Use .astype() .to() instead.
11. [Fix] reshape() without postfix const causing error when reshape with const Tensor.
12. [Enhance][Experiment] add Lstsq for least square calculation. [PR]
13. [Fix][C++] minor issue related to laterial argument passing by variables cannot properly resolved on C++
14. [Enhance] Diag now support rank-1 Tensor as input for constructing a diagonal tensor with input as diagonal elements.
15. [Enhance] Add c++ example for DMRG (Ke)
16. [Fix] Bug fixed in DMRG code and updated to the latest features.
17. [Fix] Bug in UniTensor do svd with rowrank=1 and the first rank has dimension=1.
18. [Enhance] add Scalar: abs, opeartor<, operator>, operator<=, operator>=
19. [Fix] #31 cd, cf internal swiching error for Lanczos_ER.
20. [Enhance] add specialization for Tensor iarithmetic with Sproxy.
21. [Fix] #31 cftcf Mul internal memcpy with wrong unit size.
22. [Fix] #31 type accessing now partially via Scalar, so no conflict will occur when ovld matvec() gives mismatched input and output type.
23. [Fix] Tensor / Storage set element with Sproxy or Scalar is now available.
24. [Fix] Lanczos_Gnd on f type accessing now partially via Scalar, so no conflict will occur when ovld matvec() gives mismatched input and output type.

v0.7.2
1. [Enhance] Add Tensor.set with Scalar
2. [Enhance][C++] Add Tensor initialize assignment op from initializer_list
3. [Enhance][C++] Add Storage initialize assignment op from vector & initializer list
4. [Fix] bug for set partial elements on Tensor with slicing issue.
5. [Fix][DenseUniTensor] set_rowrank cannot set full rank issue #24

v0.7.1
1. [Enhance] Finish UniTensor arithmetic.
2. [Fix] bug when using Tensor.get() accessing only single element
3. [Enhance] Add default argument is_U = True and is_vT = True for Svd_truncate() python API

v0.7
1. [Enhance] add binary op. -Tensor.
2. [Enhance] New introduce Scalar class, generic scalar placeholder.
3. [Enhance][expr] Storage.at(), Storage.back(), Storage.get_item() can now without specialization. The return is Scalar class.
4. [Enhance] Storage.get_item, Storage.set_item
5. [Enhance] Scalar, iadd,isub,imul,idiv
6. [Important] Storage.resize will match the behavior of vector, new elements are set to zero!
7. [Enhance] Scalar +,-,*,/ finished
8. [Enhance] add Histogram class and stat namespace.
9. [Enhance] add fstream option for Tofile
10. [Enhance] return self when UniTensor.set_name
11. [Enhance] return self when UniTensor.set_label(s)
12. [Enhance] return self when UniTensor.set_rowrank
13. [Fatal!][Fix] fix bug of wrong answer in Tensor slice for non-contiguous Tensor, with faster internal kernel
14. [Warning] Slice of GPU Tensor is now off-line for further inspection.
15. [Fix] bug causing crash when print non-contiguous Uint64 Tensor
16. [Fatal!][Fix] fix bug of wrong answer in Tensor set-element with slice for non-contiguous Tensor.
17. [Enhance] Network on the fly construction.
18. [Enhance] Scalar: Add on TN. TN.item()
19. [Fix] bug in Mod interanlly calling Cpr fixed.
20. [Enhance] All operation related to TN <-> Scalar
21. [Enhance] Reduce RTTR overhead.

v0.6.5
1. [Fix] Bug in UniTensor _Load
2. [Enhance] Improve stability in Lanczos_ER
3. [Enhance] Move _SII to stack.
4. [Enhance] Add LinOp operator() for mv_elem
5. [Enhance] Add c++ API fast link to cutt
6. [Enhance] Add Fromfile/Tofile for load/save binary files @ Tensor/Storage
7. [Enhance] Add linspace generator
8. [Fix] Bug in Div for fast Blas call bug
9. [Enhance] Add Tensor.append(Storage) if Tensor is rank-2 and dimension match.
10. [Enhance] Add algo namespace
11. [Enhance] Add Sort-@cpu
12. [Enhance] add Storage.numpy() for pythonAPI
13. [Enhance] add Tensor.from_storage() for python API

v0.6.4
1. [Enhance] Add option mv_elem for Tensordot, which actually move elements in input tensor. This is beneficial when same tensordot is called multiple times.
2. [Enhance] Add option cacheL, cacheR to Contract of unitensor. which mv the elements of input tensors to the matmul handy position.
3. [Enhance] optimize Network contraction policy to reduce contiguous permute, with is_clone argument when PutUniTensor.
4. [Enhance] Add Lanczos_Gnd for fast get ground state and it's eigen value (currently only real float).
5. [Enhance] Add Tridiag python API, and option is_row
6. [Enhance] C++ API storage add .back<>() function.
7. [Enhance] C++ API storage fix from_vector() for bool type.
8. [Enhance] Change Network Launch optimal=True behavior. if user order is given, optimal will not have effect.
9. [Enhance] Add example/iDMRG/dmrg_optim.py for better performace with Lanczos_Gnd and Network cache.
10. [Fix] wrong error message in linalg::Cpr
11. [Fix] reshape() on a already contiguous Tensor will resulting as the change in original tensor, which should not happened.

v0.6.3
1. [Enhance] Add Device.Ncpus for detecting avaliable omp threads
2. [Enhance] Add HPTT support on CPU permute.
3. [Internal] Build version centralize
4. [Enhance] More info for Device.
6. [Enhance] Add cytnx.__variant_info__ for checking the installed variant.

v0.6.2
1. [Fix] Bug in CUDA Matmul interface passing the wrong object bug.
2. [Enhance] Add Matmul_dg for diagonal matrix mutiply dense matrix.
3. [Enhance] Add Tensordot_dg for tensordot with either Tl or Tr is diagonal matrix
4. [Enhance] Contract dense & sparse memory optimized.
5. [example] Add iTEBD_gpu.py example
6. [Fix] Bug in CUDA cuVectordot d and f seg fault
7. [Enhance] Add cuReduce for optimized reduction.
8. [Enhance] Optimize performance for Mul_internal_cpu.
9. [Enhance] Optimize performance for Div_internal_cpu.
10. [Fix] Bug in permute of UniTensor/Tensor with duplicate entries does not return error.

v0.6.1
1. [Enhance] add Scalar class (shadow)
2. [Enhance] change default allocation from Malloc to Calloc.
3. [Enhance] change storage.raw_ptr() to storage.data() and storage.data<>()
4. [Enhance] change storage.cap to STORAGE_DEFT_SZ that can be tune.
5. [Enhance] adding Tproxy/Tproxy, Tproxy/Tensor, Tensor/Tproxy operation
6. [Enhance] Add mv_elem type for LinOp, which intrinsically omp the matvec operation.
7. [Fatal  ] Fix bug in Dot for Matrix-Vector multiplication on both GPU and CPU with complex&real float dtypes.

v0.6.0
1. [Enhance] behavior change the behavior of permute to prevent redundant copy in UniTensor and Tensor.
2. add Tensor::same_data to check if two Tensor has same storage.
3. [Enhance] the behavior of reshape in Tensor to prevent redundant copy.
4. [Enhance] behavior change all linalg to follow the same disipline for permute/reshape/contiguous
5. [Enhance] add print() in C++ API
6. [Fix] reshape() does not share memory
7. [Fix] BoolStorage print_elem does not show the first element in shape

v0.5.6a
1. [Enhance] change linalg::QR -> linalg::Qr for unify the function call
2. Fix bug in UniTensor Qr, R UniTensor labels bug.
3. Add Qdr for UniTensor and Tensor.
4. Fix minor bug in internal, Type.is_float for Uint32.
5. [Enhance] accessor can now specify with vector.
6. [Enhance] Tproxy.item()
7. Fix inplace reshape_() in new way templ. does not perform inplace operation
8. [Enhance] Tproxy operator+-/*
10. Fix bug in division dti64 becomes subtraction bug.

v0.5.5a
1. [Feature] Tensor can now using operator() to access elements just like python.
2. [Enhance] Access Tensor can now exactly the same using slice string as in python.
3. [Enhance] at/reshape/permute in Tensor can now give args without braket{} as in python.
4. [Enhance] Storage.Load to static, so it can match Tensor
5. [Major] Remove cytnx_extension class, rename CyTensor->UniTensor
6. Fix small bug in return ref of Tproxy
7. Fix bug in buffer size allocation in Svd_internal

v0.5.4a-build1
1. [Important] Fix Subtraction real - complex bug.

v0.5.4a
1. Add linalg::Det
2. Add Type.is_float
3. [Feature] Add LinOp class for custom linear operators used in iterative solver
4. enhance arithmetic with scalar Tensors
5. Add Tensor append with tensor.
6. [Feature] Add iterative solver Lanczos_ER
7. [Enhance] Tproxy +=,-=,/=,*= on C++ side
8. Add ED (using Lanczos) example.
9. Change backend to mkl_ilp64, w/o mkl: OpenBLAS
10. Change Rowrank->rowrank for CyTensor.

v0.5.3a
1. Add xlinalg.QR
2. enhance hosvd.
3. Fix bug in cytnx.linalg.Abs truncate the floating point part.
4. Add example for HOTRG
5. Add example for iDMRG
6. Add CyTensor.truncate/truncate.
7. Add linalg::Sum.
8. Complete set_elem for sparse CyTensor dispatch in binding.
9. [Important] Change Inv/Inv_ to InvM/InvM_ for matrix inverse.
10. [Important] Add Inv/Inv_ for elementwise inverse with clip.
11. [Enhance] Add str_strip for removing " ", "\t", "\r" at the end.
12. [Enhance] Accessor::() allow negative input.
13. Add GPU Pow/Pow_
14. Add random.uniform()
15. Fix bug in diagonal CyTensor reshape/reshape_ cause mismatch.
16. Add a is_diag option for convert Tensor to CyTensor.

v0.5.2a-build1
1. example/iTEBD, please modify the argument rowrank->Rowrank if you encounter error in running them.
2. Fix bug in cytnx.linalg.Abs truncate floating point part. ---> v0.5.2a-build1
3. Fix bug in mkl blas package import bug with numpy.        ---> v0.5.2a-build1

v0.5.2a
1. add Trace and Trace_ for CyTensor.
2. fix bug in Network.Launch does not return the output CyTensor
3. Add Network.PrintNet, and ostream support.
4. Add Network.Diagram() for plot the tensor network diagram (python only)
5. Add support for floating type Vectordot on GPU.
6. Fix bug in to from Anytype to ComplexFloat.
7. Add QR for CPU.
8. Add identity() and it's alias function eye().
9. Add physics namespace/submodule
10. Add physics::spin() for generating Spin-S representation.
11. Add physics::pauli() for pauli matrix.
12. Add ExpM() for generic matrix (CPU only)
13. Fix bug in python slice, and reverse range slice.
14. Enhance optional Kron padding scheme
15. Fix bug in CyTensor contract/Contract(A,B) for tensors with no common label
16. Enhance error message in Network
17. Add Min(), Max() (CPU only)
18. Fix bug in Abs.
19. Fix bug in Div i32td.
20. [Feature] Add optimal contraction order calculation in Network
21. Fix SparseCyTensor contiguous address wrong calculation.
22. Support at() directly from SparseCyTensor.
23. Add Transpose, Dagger to CyTensor. For tagged CyTensor, Transpose/Dagger will reverse the direction of all bonds.
24. Add xlinalg.Svd, xlinalg.Svd_truncate support for tagged CyTensor.
25. Fix redundant print in optimal contraction order
26. Add CyTensor.tag() for DenseCyTensor (regular type) directly convert to CyTensor with direction (tagged type)
27. Add SparseCyTensor.at (currently only floating point type)
28. SparseCyTensor.ele_exists.
29. SparseCyTensor.Transpose, Conj.
30. Symmetry.reverse_rule, Bond.calc_reverse_qnums
31. Fix Tensor.numpy from GPU bug.
32. Fix Tensor.setitem/getitem pybind bug.
33. SparseCyTensor.get_elem/set_elem (currently floating type only (complex))
34. Add xlinalg::ExpH, xlinalg::ExpM, xlinalg::Trace (ovld of CyTensor.Trace)
35. support Mul/Div operation on SparseCyTensor
36. Add Tensor.flatten();
37. Add Network.Savefile. Network.PutCyTensors
38. [Feature] Tensor can now use unify operator[] to get and set elements as python API
39. fix ambiguous error message in Tensor arithmetic.
40. fix bug in xlinalg::Svd
41. fix bug in physics::pauli
42. fix bug in CyTensor.set_label checking element.
43. Add xlinalg::Hosvd (currently CyTensor only)
44. change argument of init CyTensor rowrank->Rowrank
45. Add PESS example
46. Add support for Norm to generic rank-N Tensor
47. Add @ operator in python API for shorthand of linalg::Dot
48. Add DMRG example
49. C++ API can now have accessor.size() < rank()
50. Remove redundant output of Inv.
51. Add Pow, Pow_ for CyTensor.
52. Add Symmetry.Save/Load
53. Symmetry/Tensor/Storage/Bond/CyTensor Save/Load re-invented for more simple usage

v0.5.1a
1. add Norm() for CPU and GPU, add to call by Tn
2. add Dot() for CPU and GPU, with unify API for Vec-Vec/Mat-Vec/Mat-Mat/Ten-Vec product.
3. add Tensor.rank()
4. [Feature] support Tensor <-> numpy.ndarray
5. add random::Make_uniform()
6. Fix bug in Svd_truncate that will change the underlying block for contiguous CyTensor.
7. Fix bug in Tensor->numpy if the underlying Tensor is non-contiguous.
8. Add Eig.
9. Add real() imag() for Tensor.
10. Enhance python API, Storage & Tensor are now iterable.
11. Fix buf in Conj and Conj_, for both C++ and python
12. Fix bug python inplace call return ID Conj_, Inv_, Exp_
13. Add Conj, Conj_ for CyTensor
14. Fix non-inplace Arithmetic for non-contiguous tensor.
15. Add [trial version] Trace.
16. Add Pow, Pow_ for cpu.
17. Add Abs, Abs_ for cpu.

v0.5.0a
1. Add .imag() .real() for Storage.
2. Add xlinalg under cytnx_extension for algebra on CyTensor
3. Add xlinalg::Svd()
4. Change linalg::Eigh() to match numpy
5. fix Diag uninitialize elemets bug
6. add linalg::ExpH()
7. add random::Make_normal()
8. add iTEBD example for both C++ and python @ example/iTEBD

v0.4
1. remove Otimes, add Kron and Outer
2. Add Storage append, capacity, pre-alloc 32x address
3. Tensor can now allow redundant dimension (e.g. shape = (1,1,1,1,1...)
4. Add Storage.from_vector, directly convert the C++ vector to Storage
5. Add more intruisive way to get slices for Tensor in C++, using operator[]
6. Add Tensor.append for rank-1 Tensor
7. Add Exp() Expf() Exp\_() Expf\_()
8. Change UniTensor to CyTensor
9. Guarded CyTensor, Bond, Symmetry and Network class with cytnx_extension namespace (cytnx_extension submodule in python).