FrontISTR v4.4 Release Notes¶

Introduction to the Minor Version Update (Ver.4.4)¶

15th FrontISTR Study Group (Features / Examples / Formulation / Program Description: "Thermal Stress Analysis / Elasto-Plastic Analysis")

PExProCS, LLC. Kazuya Goto http://www.pexprocs.jp/

Overview of the Minor Version Update¶

• Fixes related to FrontISTR
  • I/O
  • Refiner
  • Analysis
    • Elements
    • Materials
    • Linear solvers
• Fixes related to the partitioner
  • I/O
  • Distributed mesh generation

Figure: Flow of analysis with FrontISTR (mesh → partitioner → distributed mesh → FrontISTR parallel computation → analysis results, and mesh → FrontISTR serial computation → analysis results).

FrontISTR: I/O¶

• Speeding up file reading
  • SECTION data
  • DLOAD data
  • FIXTEMP data
• Change in output when the 781 shell element is used
  • Change 6 degrees of freedom of DISPLACEMENT to DISPLACEMENT 3 + ROTATION 3
  • ROTATION is separated into an optional output
• Removal of debug messages

FrontISTR: Refiner¶

• Fixes related to !EQUATION
  • Fixed a defect where it was not refined correctly when a constant is set on the right-hand side

FrontISTR: Analysis¶

• OpenMP parallelization of the element loop
  • Stiffness matrix computation and stress computation now support thread parallelism
  • Hybrid parallelism performance has improved
• Fixes related to frequency response analysis
  • Fixed defects in the velocity and acceleration computations at nodes

FrontISTR: Analysis¶

• Fixes related to thermal stress analysis
  • When going through !SOLUTION, TYPE=STATIC
  • For hexahedral first-order elements, change incompatible elements to B-bar elements
• Fixes related to pressure normal-direction update
  • When going through !SOLUTION, TYPE=NLSTATIC
• Fixes to the contact stiffness matrix

FrontISTR: Elements (1/2)¶

• Fixes related to B-bar elements (solid elements)
  • Implementation for the Total Lagrangian case
• Fixes related to shell elements
  • Upper limit of the number of layers for layered shells: 15 → no limit
  • Added !DLOAD to the 33 shell element

FrontISTR: Elements (2/2)¶

• Fixes related to truss elements
  • Fix to the tangent stiffness matrix in finite deformation
  • Added GRAV to !DLOAD for trusses
  • If a diagonal component becomes 0, change the diagonal component to 1
  • Support for the refiner in trusses
  • Support for output in UCD format
• Fix to memory corruption
  • When shells, beams, trusses, and solids are all mixed

FrontISTR: Materials¶

• Fixes related to the orthotropic elastic material
  • Previously, the orthotropic elastic material was supported only when !SOLUTION, TYPE=NLSTATIC
  • → The orthotropic elastic material is now also supported when !SOLUTION, TYPE=STATIC
  • Fixed a bug in the input of orthotropic directions set in SECTION

FrontISTR: Linear Solvers (1/5)¶

• Implemented the explicit degree-of-freedom elimination method in the multi-point constraint (MPC) treatment
  • Convergence of the iterative solvers has improved
  • Switched as follows by the value of the !SOLVER parameter MPCMETHOD

• Since it cannot be computed correctly with the previous domain partitioning data, it is necessary to re-partition with the partitioner updated this time

FrontISTR: Linear Solvers (2/5)¶

• Made it possible to use the iterative method in contact analysis by the Lagrange multiplier method (experimental)
  • CG, BiCGSTAB, GPBiCG, GMRES are available
  • When performing domain partitioning, it is necessary to specify CONTACT=AGGREGATE in the partitioner's control file

FrontISTR: Linear Solvers (3/5)¶

• Estimation of matrix condition numbers by the 3x3 CG and GMRES solvers
  • Condition number estimation is performed every iteration count specified by ESTCOND in the !SOLVER parameter
  • To use it, specify the --with-lapack option in setup.sh
• Added a divergence check to the 3x3 CG solver
• Reuse of the preconditioner setup information when using the 3x3 solver
  • Analyses that repeatedly solve the same matrix, such as linear dynamic analysis and modal analysis, have been sped up

FrontISTR: Linear Solvers (4/5)¶

• Added an interface to the external AMG preconditioner library (ML) for the 3x3 solver
  • Available with the !SOLVER parameter PRECOND=5
  • To use it, specify the --with-ml option in setup.sh

FrontISTR: Linear Solvers (5/5)¶

• Fixes to the matrix dump function
  • Fixed the indices of the output matrix data
• Memory reduction of the 3x3 ILU preconditioner
  • Fixed the places that held diagonal components twice
• Added the 4x4 CG solver
• Added the 6x6 CG solver

Partitioner: I/O¶

• Speeding up file input
  • Added hash tables to the search algorithm
    • Node groups, element groups, and face groups
  • Added a cache structure to the search loop
• Extension of the log output function
  • Added the command-line option -v for detailed log output
  • MPI parallel: outputs the source process number
  • OpenMP parallel: outputs the source thread number

Partitioner: Domain Partitioning¶

• OpenMP parallelization of the distributed mesh creation loop
  • In addition to the conventional MPI parallelism, it also supports thread parallelism
• Change in the partitioning method for meshes containing MPC
  • Support for the explicit degree-of-freedom elimination

Bug Reports, etc., Recorded on the Internal Bulletin Board of the FrontISTR Study Group¶

• Typo in an if statement in element.f90
  • Fixed
• Real numbers in ElasticLinear.f90 were written in single-precision notation
  • Fixed to double-precision notation
• Integers were mixed into real-number arithmetic in static_LIB_beam.f90
  • Unified to real numbers

Scheduled Release of Ver.4.4¶

• Release date: January 2015
• Version name: Ver.4.4
• Release site: http://www.multi.k.u-tokyo.ac.jp/FrontISTR/