here - The Reiher Research Group
Contents
1. Note name defines the name under which the expectation values are stored on the resultfile Computes lopiiope p i 1 L 7 t 1 L Nup Ndown Nup Ndown cdag_up cdag_down c_up c_down cdag_up Ndown c_up Ndown cdag_down Nup c_down Nup cdag_up cdag_ down c_up c_down cdag_up c_down and cdag_down c_up as op and op are recognized by the program Available for 2u1 pg only Note name defines the name under which the expectation values are stored on the resultfile 6 3 QCMaquis tools QCMAQUIS comes with several tools that allow for example further manipulation of the MPS or to acquire additional wave function analysis information The tools det2mps_ symmetry 28 mps2ci and mps_transform _ pg will be briefly described in the following These tools are provided in the my Molcas build qcmaquis bin folder Table 8 Overview of QCMAQUIS tools tool description mps_transform _pg det2mps_symmetry mps_overlap_symmetry _pg This tool allows for a transformation of an su2ui pg MPS wave function to 2u1 pg symmetry Command line mps_transform _pg chkpfile Note for an su2u1 pg MPS with S gt 0 2u1i pg MPSs for all S components are generated This tool generates determinants based on the CI DEAS procedure 11 and inserts them in an MPS from which a new DMRG calculation can be started Starting from such an MPS is likely to improve convergence behaviour and i2. 10 100 amp Grid_it all 33 7 2 Example file for QCMaquis DMRG CASCI 8 8 active space dependent parameters L 8 nelec 8 target symmetry spin and spatial of the wave function 2 spin 0 singlet totally symmetric point group irrep spin 0 irrep 0 parameter to control the actual DMRG calculation nsweeps 8 max_bond_dimension 256 conv_thresh le 6 initialization procedure init state default technical parameters symmetry su2ulpg integral cutoff 1e 40 truncation_initial 1e 50 truncation_final le 7 chkpfile chkp h5 resultfile result h5 integral_file FCIDUMP storagedir scratch S USER boundaries LATTICE orbitals lattice_library coded MODEL quantum_chemistry model_library coded all expectation value calculations required for entropy measures MEASURE ChemEntropy 34 References 1 S Keller M Dolfi M Troyer M Reiher arXiv 1510 02026 An efficient matrix product operator representation of the quantum chemical Hamiltonian N S Battaglia A Muolo S Keller S Knecht and M Reiher in preparation wo S Keller and M Reiher in preparation a Y Ma S Keller C Stein S Knecht R Lindh M Reiher in preparation on www molcas org D P J Knowles N C Handy Comp Phys Commun 1989 54 75 A determinant based full configuration
3. 12 Setting p a build folder 2 2 05 4s e044 6 4 26 Se ee hoa ww 6 ela Ce ov ST ee R ee oS ee AE eH ES 7 ole Building ond mstala Hom e oe ses oe bee eee eee ee PS 8 3 1 5 Setting up the runtime environment sooo e a e e e 8 3 2 QCMAQUIS as external module of MOLCAS aaa a 9 ool Dopnload MOLGAS o eoo ka ec kd eoa sda RT R dees 9 3 2 2 Switch to the MOLCAS DMRG branch 10 3 2 3 Initialization of git submodul s ss ses sc ccce ec 2445 54 4 4 11 324 perme up a build folder s s s s gor ars R a r oa R RT T LES RS 11 gan Conneurshiom pe kk de e aukon ae aiig OS eRe Oe E ide TR a 11 346 B ilkding and instalation e 2 4 ze cta sota diire HERS 13 3 2 7 Setting up the runtime environment a sooo e a e e 13 3 3 Supported operating systems and compiler environments 14 3 4 First tests and verification of the installation 14 4 Maintenance 16 Al New versions and patches 6 fen hyd ee He POOH OS BREE 6 Oe Re ES 16 iii A2 Reporting bugs and user SUpport lt s See dee ee eee ee be Ew N 16 General Considerations for Running a QCMaquis DMRG Calculation 17 5 1 Memory management and memory requirements soso ooo 17 5 2 Files required and written by QCMAQUIS aaa 17 5 3 Typical workflow for a DMRG calculation including a post processing analysis 18 Input Keywords 19 6 1 Keywords and Options for MOLCAS aoaaa 00 tee 19 61 1 Gompulsory keywords spo k em or pe p eoep
4. Adrian Kantian in the group of Thierry Giamarchi For further information on the ALPS project please visit alps comp phys org Refer to the original ALPS MPS paper M Dolfi B Bauer S Keller A Kosenkov T Ewart A Kantian T Giamarchi M Troyer Comp Phys Commun 2014 12 3430 doi 10 1016 j cpc 2014 08 019 ALPS is a general open source framework for the description of strongly correlated many particle systems B Bauer et al ALPS Collaboration The ALPS project release 2 0 open source software for strongly correlated systems J Stat Mech 2011 P05001 http dx doi org 10 1088 1742 5468 2011 05 P05001 Contents 1 Introduction to the QCMaquis Software Suite 1 11 Overview and Goals 4 4 4 4 86 4 0624 See REE ERE ERS 44S 1 1 2 What features are included in the release version 1 0 1 1 2 1 QCMAQuIS standalone version sosoo o 1 122 QCMAOUI m MOLGAS 2 s aed K RR RR e RKR ba add 2 1 3 Organization of this document lt 2 44444564 245 a 2 2 Software Requirements amp Registration 4 21 Prerequisites e se roce na eani Eon diou KR Ee ee e ee Ee SESE 4 22 RECO e a ee RRS Se ee ee ee eS 4 22 1 QCMAQUIS standalone version s ces e dosc cre danei eR does 4 22 2 QCMAQUIS in MOLGAS o cored ropa ew h eoi eii h Ree eoru ih 5 3 Installation 6 ol QOUMAGUIS standalone version sss ep cave hd awe p ewe eS EO 6 Bid Download MAGUS ok ies ek be ee a eop ew ES Eee eee ES 6 3
5. This keyword is not available because it is not possible to restrict RASScf the excitation level between subspaces in DMRG calculations This keyword is not available because the Jacobi Davidson diagonalization is independent and can be controlled with the TIGHt ietl_jcd_tol and ietl_jcd_maxiter parameters see Section 6 2 in the RGINPUT section 6 1 4 Molcas environment variables As described in Section 3 2 5 3 QCMAQUIS is built by default with a shared memory OMP parallelization To speedup calculations the user can thus set at runtime the environment variable export QCMaquis_CPUS XX where XX specifies the number of shared memory cores to be used The default is to use a single core 21 6 2 Keywords and Options for QCMaquis In the following we describe i compulsory keywords Section 6 2 1 ii optional keywords Section 6 2 2 for QCMaquis DMRG calculations as well as iii keywords for property calculations Section 6 2 3 Most of the QCMAaquiIs keywords have default settings that guarantee convergence in the general case and are inserted automatically by the host program MOLCAS in this case A reasonable choice of default values for optional keywords is given in our example QC MAQUIS input file in Section 7 2 Caution MAQUIS has many features beyond quantum chemistry e g re lated to solid state physics Some keywords listed in the example file in Section 7 2 are therefore not explained in the followi
6. and installation After a successful configuration type make or make j8 to compile MOLCAS in parallel on 8 cores and install all its components in the build folder my Molcas build In the same folder QCMAQUIS as well as the required Boost and ALPS libraries will be downloaded and installed respectively The installation process thus requires a working internet connection 3 2 7 Setting up the runtime environment After having successfully passed the QCMAQUIS installation step as indicated by CMake messages like xxx Built target alps xxx Built target qcmaquis and xxx Installation of QCMaquis ALPS and Boost was successful adjust your runtime environment variables assuming a bash environment as follows export PATH path to my Molcas build alps bin PATH export PATH path to my Molcas build qcmaquis bin PATH export PYTHONPATH path to my Molcas build alps lib PYTHONPATH export PYTHONPATH path to my Molcas build qcmaquis lib python pyeval PYTHONPATH export PYTHONPATH path to my Molcas build qcmaquis lib python PYTHONPATH On Linux systems x86_64 set in addition export LIBRARY_PATH path to my Molcas build alps lib LIBRARY_PATH export LD_LIBRARY_PATH path to my Molcas build alps lib LD_LIBRARY_PATH whereas on Mac OS X set 13 export LIBRARY_PATH path to my Molcas build alps lib LIBRARY_PATH export DYLD_LIBRARY_PATH LIBRARY_PATH DYLD_LIBRARY_PATH Full instructions ready for copy p
7. configuration steps for the most popular compiler suites GNU Section 3 1 3 1 and Intel Section 3 1 3 2 respectively How to setup and use a shared memory OMP installation of QCMAQUIS is described in Section 3 1 3 3 3 1 3 1 Configuration with the GNU compiler suite To configure QCMAQUIS with the GNU compiler suite type FC gfortran CC gcc CXX gt cmake DQCM_standalone O0N path to my QCMaquis src where we assumed that the Intel Math Kernel Library MKL is available recommended option If the MKL libraries are not available QCMAQUIS will search for other suitable math libraries installed on the operating system If none are found the configuration step will stop with an appropriate message 3 1 3 2 Configuration with the Intel compiler suite To configure QCMAQUuIS with the Intel compiler suite including MKL type FC ifort CC icc CXX icpc cmake DQCM_standalone 0N path to my QCMaquis src 3 1 3 3 Shared memory OMP parallel configuration By default QCMAQUIS is built as shared memory OMP parallelized version which should work with either compiler suite GNU or Intel In order to exploit the shared memory OMP parallelization of QCMAQUIS the user is strongly encouraged to set at runtime the environment variable export OMP_NUM_THREADS XX where XX specifies the number of shared memory cores to be used The default depending on the operating system is to use a single core 3 1 4 Building and installation After a
8. e State specific and state averaged DMRG SCF calculations e Analytic gradients for state specific DMRG SCF calculations 1 3 Organization of this document In the following we list and briefly summarize the remaining sections of this document e Section 2 Software Requirements amp Registration guides through the software require ments and the registration process for QCMAQUIS e Section 3 Installation guides through the installation process for QCMAQUIS and possibly the host program MoLcas e Section 4 Maintenance summarizes general information concerning the QCMAQUIS software suite including how to ask for user support retreive future patches for the code and where to send bug reports e Section 5 General Considerations for Running a QCMAquiIs DMRG Calculation in troduces the basic workflow for QCMaquis DMRG calculations and provides details about memory requirements as well as input and output data required and generated by QCMAQUIS respectively e Section 6 Input Keywords provides in the first part a list of keywords and options to control QC M AQUIS DMRG calculation in MoLcAS The second part explains in detail all mandatory and optional QCMAQUIS keywords and their usage In addition we introduce several tools that are part of the QCMAQUIS software suite which can be used to analyze and visualize the resulting DMRG wave function s e Section 7 Examples of MOLCAS DMRG and QCMaAquis DMRG Input Files then s
9. other in the one dimensional lattice The first ordering in the output ignores the point group symmetry of the orbitals while the second version orders the orbitals within each irreducible representation and then reorders these symmetry blocks Command line fiedler py resultfile Note we recommend to use the second option This script recovers the complete QCMAQUIS input file dmrg input from a given resultfile Command line input py resultfile 31 Table 9 continued from previous page script description mutinf py Produces mutual information plots 13 given that an expectation value calculation for MEASURE ChemEntropy has been performed Command line mutinf py resultfile Note if orbital images in png format named 1 png 2 png L png with L being the number of active orbitals are present in the same folder where mutinf py is executed they can be added to the mutual information plot by providing the optional argument i to mutinf py 32 7 Examples of Molcas DMRG and QCMaquis DMRG Input Files 7 1 Example file for Molcas DMRG SCF 14 10 amp GATEWAY 2 Angstrom N 0 000000 0 000000 0 54880 N 0 000000 0 000000 0 54880 basis cc pvdz amp SEWARD amp SCF amp RASSCF DMRG RGinput UI QCMAQUIS Keywords input conv_thresh 1 0E 07 nsweeps 4 max_bond_dimension 100 endRG UI End of QCMAQuIS Keywords input inactive 00000000 RAS2 3 1103 11 0 ITER
10. to an existing local MOLCAS repository in my Molcas src that does NOT yet have a pointer to the remote repository git molcas molcas Section 3 2 1 2 3 2 1 1 Cloning a new local Molcas repository Type git clone recursive git molcas molcas my Molcas src to download a new local MOLCAS repository in the source folder my Molcas src 3 2 1 2 Adding a new remote repository to an existing local Molcas repository Type git remote add gitsrc_molcas git molcas molcas to add the additional remote repository gitsrc_molcas that hosts the MOLCAS DMRG branch to your local MOLCAS repository which we assume in the following to be located in my Molcas src To obtain all remote informations from gitsrc_molcas type next git fetch gitsrc_molcas 3 2 2 Switch to the Molcas DMRG branch QCMAQUIS is available on the master branch of MOLCAS which is the default branch that you currently reference if you git cloned a new local MOLCAS repository in the previous step Typing git branch should then yield something like master The x in front of the branch name indicates which local branch you are currently tracking In this particular case you are tracking the branch master If you added a new remote repository gitsrc_molcas to your existing local MOLCAS repository see Section 3 2 1 2 type git checkout b dmrg master gitsrc_molcas master 10 Typing next git branch should then yield something like master d
11. A quick user guide to the QCMAQUIS software suite for MOLCAS Sebastian Keller Stefan Knecht Yingjin Ma Christopher Stein and Markus Reiher ETH Zurich Laboratorium fiir Physikalische Chemie Vladimir Prelog Weg 2 CH 8093 Z rich December 5 2015 release version 1 0 We kindly request that for reproducibility reasons any use of the QCMAQUIS software suite for density matrix renormalization group DMRG calculations in MOLCAS that results in published material should cite the set up steered by settings and warm up procedures described in Check for a preprint on arXiv org to appear soon Y Ma S Keller C Stein S Knecht R Lindh M Reiher in preparation The DMRG calculations are then conducted with the software QCMAQUIS that requires a citation It is described in the following paper Check for the journal article asap on arXiv org S Keller M Dolfi M Troyer M Reiher arXiv 1510 02026 physics comp ph QCMAatis builds upon the ALPS MPS project The ALPS MPS codes implement the DMRG algorithm for variational ground and low lying excited state search as well as time evolution of arbitrary one and two dimensional models in a matrix product state representa tion They have been developed at ETH Zurich by Michele Dolfi and Bela Bauer in the group of Matthias Troyer with contributions from Sebastian Keller and Alexandr Kosenkov and at the University of Geneva by Timothe Ewart and
12. AQUIS Ton install QCMAQUIS as external module of the quantum chemistry software MOLCAS proceed to Section 3 2 3 1 QCMaquis standalone version In the following steps 3 1 1 3 1 5 we describe how to successfully build and install the QC MAQUIS software suite The installation of QCMAQUIS has been tested for different operating systems and compiler math libraries environments Their list can be found in Section 3 3 While other combinations might work equally well they are not officially supported The installation of the QCMAQUIS software suite will comprise several libraries which are automatically downloaded and installed during the QCMAQUIS build process e QCMAQUIS e Boost e ALPS All of the above libraries will be installed locally in the user defined build folder my QCMaquis build 3 1 1 Download QCMaquis Type git clone git tc gitlab ethz ch qcmaquis qcmaquis public git my QCMaquis src to download a new local QCMAQUIS repository in the source folder my QCMaquis src 3 1 2 Setting up a build folder Create a build folder my QCMaquis build note that this folder does not necessarily have to be a subfolder of my QCMaquis src and change to this new folder mkdir path to my QCMaquis build amp amp cd path to my QCMaquis build 3 1 3 Configuration Table 1 in Section 3 3 summarizes the list of tested and supported operating system and com piler combinations for the installation of QCMAQuUIS Below we will show the
13. TE Ikone SBE ERS 19 61 2 Optional keywords scs se ec sogea sra ku draper au 19 6 1 3 Inoperative keywords ooo a 19 6 1 4 Molcas environment variables lt s e o soeu seua 844 4s 21 6 2 Keywords and Options for QCMAQUIS oa aaa A Be oe 22 6 2 1 Compulsory keywords ra srota dhe ete Se ee E a 22 6 2 2 Optional keywords aoaaa a 24 6 2 3 Keywords for expectation value calculations 27 03 QCMAQUIS tOOlS gt sac a s acro eee ee E P REE ee ae Oe R 28 6 4 QCMAQUIS PYTHON scripts for wave function analysis and visualization 30 Examples of Molcas DMRG and QCMaquis DMRG Input Files 33 7 1 Example file for MoLrcas DMRG SCF 14 10 33 7 2 Example file for QCMAquis DMRG CASCI 8 8 aaa 34 iv 1 Introduction to the QCMaquis Software Suite 1 1 Overview and Goals The QCMAQUIS software suite allows for an efficient optimization of a matrix product state MPS wave function based on a second generation DMRG algorithm 1 The quantum chemical operators are represented as matrix product operators MPOs which provides the necessary flexibility to accommodate abelian and non abelian symmetries as well as the imple mentation of non relativistic and relativistic quantum chemical Hamiltonians 2 respectively in a unified framework We have implemented the special unitary group of degree 2 SU 2 in the MPO representation of the non relativistic Hamiltonian to ensure spin conservation 3 The current im
14. alone version of the QCMAQUIS software suite requires a registration at http tc gitlab ethz ch To sign up at http tc gitlab ethz ch please send an e mail to dmrg phys chem ethz ch After confirmation of your registration login to http 4 tc gitlab ethz ch and upload your local public ssh key to your gitlab profile Verify that you have been granted access to the following projects for example they should be listed on the right as projects you are enrolled at e QCMAaQutis QCMAQUIS PUBLIC e QCMAaQutIs QCMAQUIS SRC If this is not the case please send a notification to dmrg phys chem ethz ch 2 2 2 QCMaquis in Molcas The use of the QCMAQUIS software suite in MOLCAS requires apart from a valid MOLCAS developers license a registration at http tc gitlab ethz ch To sign up at http tc gitlab ethz ch please send an e mail to dmrg phys chem ethz ch After confirmation of your registration login to http tc gitlab ethz ch and upload your local public ssh key to your gitlab profile Verify that you have been granted access to the following projects for example they should be listed on the right as projects you are enrolled at e QCMAautis QCMAQUIS PUBLIC e QCMAautis QCMAQUIS SRC e MOLCAS DEV HDF5 UTILS e MOLCAS DEV DMRG INTERFACE UTILS If this is not the case please send a notification to dmrg phys chem ethz ch 3 Installation Section 3 1 describes in details the installation process for a standalone version of QC M
15. aste will also be printed on the screen and shall be copied to either your local HOME bashrc file or be executed in the MOLCAS bash session Alternative use the sourceable configuration file qcmaquis sh that is created after the build install step succeeded To do so type source path to my Molcas build qcmaquis bin qcmaquis sh 3 3 Supported operating systems and compiler environments Table 1 summarizes tested and officially supported operating system compiler math library combinations for building MOLCAS together with the QCMAQUIS software suite Note that other combinations might work equally well but they are not officially supported at present See the MOLCAS manual at www molcas org for details on supported operating systems and compiler environments of a plain MOLCAS installation Table 1 Supported operating system compiler math library combinations for the combined installation of MOLCAS and QCMAQUIS operating system compiler math library Mac OS X 10 11 El Capitan GNU 5 2 Accelerate Xcode Mac OS X 10 11 El Capitan Intel XE 15 patch 0 MKL 11 2 x86_64 Intel XE 15 patch 0 MKL 11 2 x86_64 GNU 4 8 MKL 11 1 x86_64 GNU 4 7 2 MKL 11 1 x86_64 Intel XE 13 MKL 11 1 x86_64 Intel XE 13 sp 1 MKL 11 1 x86_64 OpenMPI 1 6 5 Intel XE 13 sp 1 MKL 11 1 3 4 First tests and verification of the installation To verify that your installation of MOLCAS and QCMAQUIS was s
16. ectrons Irreducible representation of the point group symmetry of the irrep target state Note Counting starts with 0 which has to be the totally symmetric representation Total spin 2 x S of the target state for example spin spin 0 singlet spin 1 doublet spin 2 triplet L Length of lattice number of orbitals in the active space integral_file chkpfile resultfile n_ortho_states ortho_states Path and filename of the integral file for example integral file path to file FCIDUMP Path and name of the folder in which the MPS is stored Path and filename of the result file If an excited state calculation is desired the number of states the current wave function is to be orthogonalized against shall be specified here Path s and filename s of the MPS checkpoint file s that store the lower lying states to which the current MPS shall be orthogonal to 23 Table 5 continued from previous page keyword description Possible options are default thin and hf The default and thin initializations fill the initial MPS with random numbers the difference being that a singular value decomposition reduces the bond dimension to init_bond_dimension in the case of thin init_state Usage of hf generates an MPS consisting of only the determinant defined on the hf_occ card Note that the CI DEAS procedure 11 4 as invoked by dmrginit py see Section 6 4 behaves like a restart from the newly gene
17. hows two example input files for QCMAquis DMRG calculations in MOLCAS and with the QCMAQUIS standalone suite respectively 2 Software Requirements amp Registration 2 1 Prerequisites In order to install either QCMAQUIS or the developers version of the quantum chemistry package MoLcas with DMRG support through the QCMAQUIS software suite 1 4 requires the following libraries programs e Git e Python version 2 x with x gt 5 e HDF5 http www hdfgroup org HDF5 e GNU Scientific Library GSL http www gnu org software gs1 e CMake version 3 x with x gt 0 https cmake org Please make sure that these libraries programs are available and their location is visible in your PATH and LD_LIBRARY_PATH respectively Note that these libraries are NOT part of the installation package of the QCMAQUIS software suite see Section 3 for further details Warning The combined installation of MOLCAS and QCMAQUIS requires to follow the CMake installation of MOLCAS whereas a configure based installation is NOT possible A detailed step by step installation guide is provided in Section 3 2 2 2 Registration QCMAQUIS can be installed either as standalone version Section 3 1 or as external module of the quantum chemistry software MOLCAS Section 3 2 In the former case proceed with the registration as described in Section 2 2 1 else with Section 2 2 2 2 2 1 QCMaquis standalone version The use of the stand
18. ibraries are available recommended op tion If the MKL libraries are not available internal math libraries of MOLCAS can be requested with DLINALG Internal default option for LINALG in MOLCAs 3 2 5 2 Configuration with the Intel compiler suite To configure MOLCAS and QCMAQUIS with the Intel compiler suite including MKL type FC ifort CC icc CXX icpc cmake DDMRG O0N DLINALG MKL path to my Molcas srce 3 2 5 3 MPI parallel and shared memory OMP parallel configurations Installing an MPI parallel version of MOLCAS with DMRG support is possible although QC MAQUIS itself is by default shared memory OMP but not yet MPI parallelized To configure Mo cas for an MPI installation type FC mpif90 CC mpicc CXX mpiCXX cmake DDMRG 0N path to my Molcas src A shared memory OMP parallelized version of MOLCAS can be activated with the option DOPENMP ON passed to cmake during the configuration step for example FC CC CXX cmake DDMRG 0N DOPENMP 0N path to my Molcas src It should work with either compiler suite GNU or Intel but the user may want to consult the MOLCAS user manual for further information In order to exploit the shared memory OMP parallelization of QCMAQUIS which is enabled by default the user is strongly encouraged to set at runtime the environment variable export QCMaquis_CPUS XX where XX specifies the number of shared memory cores to be used The default is to use a single core 12 3 2 6 Building
19. ingular value decomposition Number of sweeps in which truncation_initial is used in the ngrowsweeps singular value decomposition 6 2 3 Keywords for expectation value calculations QCMAQUIS can in principle compute expectation values for any one or two particle oper ator that can be formulated in second quantization Table 7 comprises a list of the available property keywords in the release version of QCMAqultIs For further updates other properties please contact dmrg phys chem ethz ch The one particle reduced density matrix as well as the one particle spin density matrix are implicitly computed from the expectation values of some of the operators contained in MEASURE ChemEntropy 27 Table 7 Expectation value calculations available in the release version of QCMAQUIS keyword description MEASURE ChemEntropy MEASURE 1rdm MEASURE 2rdm MEASURE_LOCAL name op MEASURE HALE CORRELATIONS name op opa T All expectation values over the operators required to calculate the mutual information as specified in Ref 13 will be computed Please note that this is available only for SU2 symmetry Computes the one particle reduced density matrix without the additional correlators contained in the ChemEntropy measurement Computes the two particle reduced density matrix Computes wlop W i 1 L Nup Ndown and Nup Ndown are meaningful choices for op Available for 2u1 pg only
20. interaction program 7 S F Keller and M Reiher Chimia 2014 68 200 203 arXiv 1401 5497 Determining Factors for the Accuracy of DMRG in Chemistry 8 L Freitag S Knecht S F Keller M G Delcey F Aquilante T Bondo Ped ersen R Lindh M Reiher and L Gonz lez Phys Chem Chem Phys 2015 DOI 10 1039 C4CP05278A Orbital entanglement and CASSCF analysis of the RuNO bond in a Ruthenium nitrosyl complex 9 T Dresselhaus J Neugebauer S Knecht S Keller Y Ma and M Reiher J Chem Phys 2015 142 044111 arXiv 1409 1953 Self consistent embedding of density matrix renormalization group wavefunctions in a den sity functional environment 10 E D Hedegaard S Knecht J S Kielberg H J Aa Jensen and M Reiher J Chem Phys 2015 142 224108 arXiv 1502 06157 Density matrix renormalization group with efficient dynamical electron correlation through range separation 11 O Legeza J S lyom Phys Rev B 2003 68 195116 Optimizing the density matrix renormalization group method using quantum information entropy 35 12 G Barcza Legeza K H Marti M Reiher Phys Rev A 2011 83 012508 Quantum information analysis of electronic states of different molecular structures 13 K Boguslawski P Tecmer O Legeza M Reiher J Phys Chem Lett 2012 3 3129 Entanglement measures for single and multireference correlatio
21. mrg master The x in front of the branch name indicates which local branch you are currently tracking In this particular case you successfully switched to track the development branch dmrg master 3 2 3 Initialization of git submodules Type git submodule update init recursive to initialize and possibly download the necessary git submodules of MOLCAS 3 2 4 Setting up a build folder Create a build folder my Molcas build note that this folder does not necessarily have to be a subfolder of my Molcas src and change to this new folder mkdir path to my Molcas build amp amp cd path to my Molcas build 3 2 5 Configuration Table 1 in Section 3 3 summarizes the list of tested and supported operating system and compiler combinations for the simultaenous installation of MOLCAS and QCMAaquIS Below we will show the configuration steps for the most popular compiler suites GNU Section 3 2 5 1 and Intel Section 3 2 5 2 respectively How to setup an MPI parallel MOLCAS installation is described in Section 3 2 5 3 Note We strongly recommend to configure QCMAQquIS and MOLcaAs with the Intel Math Kernel Library MKL to ensure the best numerical performance 11 3 2 5 1 Configuration with the GNU compiler suite To configure MOLCAS and QCMAQUIS with the GNU compiler suite type FC gfortran CC gcc CXX g cmake DDMRG ON DLINALG MKL path to my Molcas src where we assumed DLINALG MKL that the MKL l
22. n 5 3 may be skipped 5 1 Memory management and memory requirements The memory layout of QCMAQUISs is designed for large multinode architectures and therefore the memory consumption on a single node is not consequently minimized Note that if you run a MOLCAS QCMAaAaqulis DMRG calculation the memory consump tion and memory allocation of QCMAQUIS is entirely disconnected from the host program MoLcas This in turn means that the amount of core memory assigned to the host program should be carefully chosen For example DMRG SCF calculations for large active spaces will not require a considerably larger amount of memory in the RASSCF module apart from memory for reduced one and two particle density matrices that scale approximately with the 4 act number of active orbitals Nac as N2 and N4 respectively than regular CASSCF calcula tions with modest sized active spaces The reason being that in the former case we do not need to reserve memory for CI vector s expansions which grow roughly factorily with the number of electrons and Nact 5 2 Files required and written by QCMaquis QCMAQUIS requires only two files to start a calculation i an input file named for example dmrg input that includes all keywords and options as specified in Section 6 2 and ii an integral file in the FCIDUMP format as described in Ref 6 The latter can be produced for example with the Motcas QCMAQUuIS host program driver described in Ref 4 Diffe
23. n effects 14 G Moritz M Reiher J Chem Phys 2007 126 244109 Decomposition of density matrix renormalization group states into Slater determinant basis 15 J Rissler R M Noack S R White Chem Phys 2006 323 519 Measuring orbital interaction using quantum information theory 36
24. ng and are not to be changed if a quantum chemical calculation is desired 6 2 1 Compulsory keywords The keywords in Table 5 have to be set for every DMRG calculation since they may crucially affect the accuracy of the final result Their choice depends for example on the molecule under consideration do you expect strong static electron correlation and or dynamical correlation to play a major role the nature of the reference orbitals Hartree Fock orbitals natural orbitals of some kind the size of the active space and many other aspects Note We strongly encourage any new user of QC MAQUIS or QC MAQUIS in MoLcas to carefully read first Ref 7 which gives a detailed intro duction to DMRG calculations in quantum chemistry Further quantum chemical DMRG sample calculations starting from different computational setups are discussed for example in Refs 4 8 9 10 Some of the compulsory keywords listed in Table 5 are indeed automatically set if you run a QCMAQuIs DMRG calculation through the QCMAQUIS host program driver which is the case for MOLCAS QC MAQulIs DMRG calculations In this case skip the upper part of Table 5 and proceed immediately to the lower part marked by Keywords NOT set by the host program MOLCAS 22 Table 5 Compulsory keywords to be set in all QC NM AQUIS DMRG calculations keyword description Keywords set by the host program MOLCAS nelec Total number of el
25. opied to either your local HOME bashrc file or be executed in the QCMAQUIS bash session Alternative use the sourceable configuration file qcmaquis sh that is created after the build install step succeeded To do so type source path to my QCMaquis build qcmaquis bin qcmaquis sh 3 2 QCMaquis as external module of Molcas In the following steps 3 2 1 3 2 7 we describe how to successfully build and install the MOLCAS program together with the QCMAQUIS software suite The installation of QCMAQUIS has been tested for different operating systems and compiler math libraries environments Their list can be found in Section 3 3 While other combinations might work equally well they are not officially supported The installation of the QCMAQUIS software suite within MOLCAS will comprise several li braries which are automatically downloaded and installed during the MOLCAS build process provided that DMRG support within MOLCAS is requested by the user The list of external libraries comprises e QCMAQUIS e QCMAQUIS driver utils e Boost e ALPS All of the above libraries will be installed locally in the user defined build folder my Molcas build of MOLCAS 3 2 1 Download Molcas If you already have a local MOLCAS repository in my Molcas src that points to the remote repository git molcas molcas skip this section otherwise choose either to download a new local MOLCAS repository Section 3 2 1 1 or add a new remote repository
26. ote Replaces max_bond_dimension which does NOT need to be specified in this case Example nsweeps 3 sweep_bond_dimensions 300 400 500 Adjusts the maximal bond dimension of the MPS produced by the CI DEAS procedure 11 4 Sets the energy convergence threshold in Hartree If the lowest energy from the previous sweep differs from the lowest energy of the current sweep by less than conv_thresh the DMRG calculation stops Note Requires to set also truncation_final and iet1l_jcd_tol Numerical format xe y with x and y being integers Example conv_thresh 1e 6 Convergence threshold for the Jacobi Davidson diagonalization Numerical format xe y with x and y being integers Example ietl_jcd_tol 1e 6 Maximum number of iterations in the Jacobi Davidson diagonalization 25 Table 6 continued from previous page keyword description truncation_initial truncation_final measure_each symmetry chkp_each If during the ngrowsweeps the sum of the discarded singular values for m retained states is lower than the value defined here more block states will be discarded until the discarded sum increases to truncation_initial Numerical format xe y with x and y being integers Example truncation_initial 1e 6 If during the nmainsweeps the sum of the discarded singular values for m retained states is lower than the value defined here more block states will be discarded until the disca
27. plementation of QCMAQUIS allows for efficient full Cl type calculations of active space sizes beyond capabilities CAS 18 18 of standard CI approaches The QCMAQUIS software suite is also available 4 within the framework of MOLcas 5 where we have implemented a state specific and state average DMRG self consistent field DMRG SCF algorithm the possibility to include solvent effects in DMRG calculations and provide analytic gradients for state specific calculations The latter enables structure opti mization within the QCMAquIs DMRG framework Advice The break even point wrt computational cost for a DMRG CI SCF calculation compared to a traditional CAS CI SCF calculation is approximately reached for a CAS 14 14 space For active spaces smaller than CAS 14 14 we recommend to choose a traditional MoLcas CAS approach 1 2 What features are included in the release version 1 0 1 2 1 QCMaquis standalone version The release version 1 0 of the QCMAQUIS software suite includes e optimization of spin adapted SU 2 MPS wave functions with the DMRG algorithm e non relativistic and scalar relativistic quantum chemical Hamiltonians e calculation of excited states e a python tool set to analyze the MPS wave function and its quantum entanglement 1 2 2 QCMaquis in Molcas The release version 1 0 of the QCMAQUIS software suite in MOLCAS supports e DMRG CI and DMRG SCF calculations w wo reaction field e g PCM
28. rated CI DEAS MPS default in Molcas init_state default Keywords NOT set by the host program MOLCAS Maximum number of renormalized states commonly referred to max_bond_dimension as m value or virtual bond dimension kept during each microiteration step of a forward or backward sweep Maximum number of DMRG sweeps Please be aware that nsweeps nsweeps sets the number of combined forward and backward sweeps Thus the actual number of sweeps is 2 x nsweeps 6 2 2 Optional keywords The keywords summarized in Table 6 may be exploited by the more experienced user but can be safely ignored by those who just want to get started They may affect the convergence and accuracy of the final result though For the inexperienced user however we advise to not change these settings and accept the default values provided by QCMAQUIS 24 Table 6 Optional keywords for QCMAQUIS calculations keyword description orbital_order sweep_bond_ dimensions init_bond_dimension conv_thresh ietl_jcd_tol ietl_jcd_maxiter Manual ordering of the orbitals along the one dimensional lattice The order has to be entered as a string of comma separated orbital numbers We recommend the Fiedler ordering based on the mutual information 4 12 which can be obtained by means of the python script fiedler py see Section 6 4 default orbital order 1 2 3 4 5 6 Sets max_bond_dimension for each sweep separately N
29. rd description RGINput XXX ENDRG CIONLY FCIDUMP SOCCupy RGInput marks the beginning of the QCMAQUIS parameter control section while ENDRG marks its end Any QCMAaqulis internal keyword can be forwarded directly to QCMAQUIS by replacing the XXX with a list of keywords each keyword in a new line A list of compulsory optional and property calculation keywords for QCMAQUIS are described in Sections 6 2 1 6 2 3 Notable exceptions are the active space and wave function symmetry dependent parameters L nelec spin irrep that will be forwarded automatically by MOLCAS This keyword disables orbital optimization A single DMRG CI calculation is performed instead This keyword disables orbital and wave function optimization that is input information and transformed integrals are written to a formatted integral file named FCIDUMP see Ref 6 for a detailed description of the format which can be used in subsequent QCMaquis DMRG calculations Initial electronic configuration for the calculated state s For the calculation of a single state this is equivalent to the hf_occ card in the QCMAQUIS input see Section 6 2 The occupation is inserted as a string of blank space separated aliases of occupations of the active RAS2 orbitals with the aliases 2 full u up d down 0 empty 20 Table 4 Inoperative keywords in the RASSCF section when DMRG is active keyword description
30. rded sum increases to truncation_final Numerical format xe y with x and y being integers Example truncation_final 1e 6 Tells the program to compute the expectation values every 2 X measure_each sweeps Defines the total symmetry group Default is the combined spin SU2 and point symmetry group PG su2uipg where the pg suffix should be omitted for better performance if the molecule is Cl symmetric For test purposes it is possible to switch off spin adaptation again with or without point group symmetry 2u1 pg In the latter case the keywords spin and nelec see Section 6 2 1 have no meaning Instead u1_total_charge1 and u1_total_charge2 corresponding to the number of up and down electrons have to be specified Tells the program to update the checkpoint file every 2 x chkp_each sweeps 26 Table 6 continued from previous page keyword description Occupation of the starting orbitals e g Hartree Fock occupation to be entered as a comma separated string of occupation aliases The aliases are defined as follows 4 full 3 up 2 down 1 empty This information has to be entered in case of hf as hf_occ init_state and for the CI DEAS procedure 11 4 as invoked by dmrginit py example hf_occ 4 4 4 2 2 1 for a CAS 8 6 triplet state setup where the unpaired electrons are located in orbital 4 and 5 Number of sweeps in which truncation_final is used in the nmainsweeps s
31. rent types of reference orbitals may be used as starting orbitals for a QCMaAquis DMRG calcula tion 17 QCMAQUIS produces two types of files i a result file in which all information on e g the energies and expectation values is stored and ii a checkpoint folder which contains the matrix product state MPS wave function The checkpoint folder is required for a later restart of the calculation while the tools and the analysis scripts described in Sections 6 3 and 6 4 respectively require either a result file or a checkpoint folder or both All data is stored in the hdf5 format 5 3 Typical workflow for a DMRG calculation including a post processing analysis The usual workflow to set up run and analyze a DMRG calculation proceeds as follows e prepare an FCIDUMP integral file 6 using the MoLcaAs QCMAQuIS host program driver 4 starting from a set of previously computed reference orbitals e prepare an input file dmrg input starting for example from the template input file provided in Section 7 2 and adjust all molecular system and wave function specific parameters for example nelec spin irrep L see Sections 6 2 1 6 2 3 for a list of all compulsory and optional input arguments e run the DMRG calculation with dmrg dmrg input tee out e compute expectation values with dmrg meas dmrg input e analyze the results using the PYTHON tools provided with the QC MAQUIS package see Sections 6 3 and 6 4
32. respectively for a list of utility programs and scripts 18 6 Input Keywords Caution Be aware of underscores in some of the QCMAQUIS or MoLrcas keywords They might get lost when you copy paste keywords from this pdf to your input file 6 1 Keywords and Options for Molcas 6 1 1 Compulsory keywords This section describes the QCMAQUIS input to the RASSCF program within the MOLCAS program system The minimal input to run a DMRG SCF calculation with QCMAQUIS in the framework of the RASSCF program is amp RASSCF DMRG where the keyword DMRG is compulsory to activate the DMRG functionality in MOLCAS 6 1 2 Optional keywords Table 3 summarizes optional keywords under the RASSCF input card that allow to control the type of DMRG calculation that is DMRG SCF DMRG CI or just a dump of information input one and two electron integrals to perform a DMRG calculation outside of MOLCAS The detailed input control for QCMAQUIS in MOLCAS is designed such that parameters between the RGINput and ENDRG section of the RASSCF input card only affect QCMAQUIS that is they have no effect within the RASSCF module of MOLCAS These parameters are described in Section 6 2 6 1 3 Inoperative keywords Table 4 comprises a list of keywords under the RASSCF input card which are have no meaning when a DMRG calculation is requested 19 Table 3 Optional keywords under the RASSCF card to control the DMRG calculation keywo
33. s less prone to get stuck in local minima The current implementation is described in Ref 4 The number of determinants will be chosen such that the maximal bond order at any site does not exceed the value set in init_bond_dimension The new MPS will be stored in the checkpoint folder specified in chkpfile Command line det2mps_ symmetry dmrg input Note symmetry must equal the total symmetry specified in the input file dmrg input This tool calculates the overlap between two MPS and Oz respectively according to 911921 Command line mps_overlap_ symmetry _pg chkpfile_1 chkpfile_2 Note symmetry must equal the full symmetry specified in the input file qmrg input Table 8 continued from previous page tool description mps2ci_2u1 pg Given a text file determant_list txt containing a list of determinant strings this tool calculates the Cl coefficients of the respective determinants 14 The determinant strings have to encode the occupation of the orbitals as described for the hf_occ keyword 4 full 3 up 2 down 1 empty Command line mps2ci_2ui pg chkpfile determinant_list txt Note with the conversion tool mps_transform _pg this analysis becomes possible also for MPS of the full symmetry su2u1 pg 6 4 QCMaquis PYTHON scripts for wave function analysis and visu alization The python scripts of QCMAQUIS are helpful to analyze and
34. successful configuration type make or make j8 to compile QCMAQUIS in parallel on 8 cores and install all its components in the build folder my QCMaquis build In the same folder BOOST and ALPS will be downloaded and installed respectively The installation process thus requires a working internet connection 3 1 5 Setting up the runtime environment After having successfully passed the QCMAQUIS installation step as indicated by CMake messages like xxx Built target alps xxx Built target qcmaquis and xxx Installation of QCMaquis ALPS and Boost was successful adjust your runtime environment variables assuming a bash environment as follows export PATH path to my QCMaquis build alps bin PATH export PATH path to my QCMaquis build qcmaquis bin PATH export PYTHONPATH path to my QCMaquis build alps lib PYTHONPATH export PYTHONPATH path to my QCMaquis build qcmaquis lib python pyeval PYTHONPATH export PYTHONPATH path to my QCMaquis build qcmaquis 1lib python PYTHONPATH On Linux systems x86_64 set in addition export LIBRARY_PATH path to my QCMaquis build alps lib LIBRARY_PATH export LD_LIBRARY_PATH path to my QCMaquis build alps lib LD_LIBRARY_PATH whereas on Mac OS X set export LIBRARY_PATH path to my QCMaquis build alps lib LIBRARY_PATH export DYLD_LIBRARY_PATH LIBRARY_PATH DYLD_LIBRARY_PATH Full instructions ready for copy paste will also be printed on the screen and shall be c
35. the user s release version of QCMAQUIS 4 2 Reporting bugs and user support The QCMAQUIS program suite is distributed to the MOLCAS community with no obligations on the side of the authors The authors thus take no responsibility for the performance of the code nor for the correctness of the results This distribution policy gives the authors no responsibility for problems experienced by the users when using the QCMAQUIS program in MOLCAS Bugs suggestions for improvements are to be reported as issues on the gitlab server Please open an issue concerning e QCMAQUIS on the project page QCMaquis QCMaquis public e QCMAQUIS in MOLCAS on the project page molcas dev dmrg interface utils Any issue will be dealt with by one of the authors although no responsibility on the prompt ness of response is given In general serious bugs that have been reported and fixed will lead to a new patch of the QCMAQUIS program announced and distributed from the QCM AQquIS homepage 16 5 General Considerations for Running aQCMaquis DMRG Calculation Running a QCMAQuIs DMRG calculation either DMRG CI or DMRG SCF within the framework of MOLCAS involves the QCMAQUIS host program driver described in Ref 4 In doing so MOLCAS DMRG calculations are easily accessible since they are integrated in the RASSCF module of MoLcAs Hence if the user considers to run a MOLCAS DMRG calculation rather than doing a standalone QCMAQquIs DMRG calculation sectio
36. uccessful type in your build folder my Molcas build 14 molcas verify qcmaquis A message like Verification has been completed indicates that all tests passed correctly and your installation is good to go for production work Note that all test inputs can be found in my Molcas build test qcmaquis They may also serve as sample inputs for your actual calculation Table 2 comprises a short summary of the seven tests for QCMAQUIS covering different computational aspects within the MOLCAS framework Table 2 List of test inputs that are stored in my Molcas build test qcmaquis file type details UU Input State specific introduces the DMRG keyword under the RASSCF card 002 input State average two state So S1 DMRG SCF of N 003 input DMRG CI dynamic list of renormalized states m 004 input State specific structure optimization of the ground state of dioxetanone 005 input State specific structure optimization of the S state of N 006 input PCM model DMRG SCF for a non equilibrium state 007 input State specific DMRG SCF with manual orbital ordering 15 4 Maintenance 4 1 New versions and patches New versions of the QCMAQUuIS and or QC MAqulis driver software suite as well as possible bug fixes will be announced on the QCMAQUIS homepage The latter patches will be made available as git commit s on the QCMAQUIS repository at http tc gitlab ethz ch that can be directly applied git merge d to
37. visualize the results of a DMRG calculation Their usage should be evident from the documentation strings in the PYTHON files However those that are most frequently used will be briefly explained here All scripts except dmrginit py take the QCMAQUIS output file resultfile as input They are located in the folder my Molcas build qcmaquis 1lib python pyeval Table 9 Overview of QCMAQUIS PYTHON analysis and visualization scripts script description Plots the energy for each microiteration Command line sweeps py sweeps py resultfile Note use this tool to check the convergence wrt the number of sweeps 30 Table 9 continued from previous page script description dmrginit py fiedler py input py Starts a QCMaaquis DMRG calculation with max_bond_ dimension 200 and nsweeps 2 measures the entropy information 15 13 from this unconverged calculation and based on this generates a new MPS with the CI DEAS procedure according to all settings specified in the input file dmrg input We recommend to use this script for the preparation of calculations for active spaces that are larger than those that can be handled with traditional CAS methods Command line dmrginit py dmrg input Optimizes the ordering based on entropy information as proposed in Ref 12 The current implementation is described in Ref 4 The ordering ensures that highly entangled orbitals are close to each
Download Pdf Manuals
Related Search