SOLVTYP

Bulk Data Entry Defines the solver type to be used for static and dynamic analysis.

Format

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
SOLVTYP SID SOLVER              
Continuation line for SOLVER = PCG, MIXED and AUTO
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
  PCON MAXIT ITOL TOL          
Continuation line for SOLVER = MUMPS
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
  ORDM                

Example

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
SOLVTYP 4 PCG              
  FAI                

Alternate Example

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
SOLVTYP 1 MUMPS              
  PORD                

Definitions

Field Contents SI Unit Example
SID Unique set identification number.

No default (Integer > 0)

 
SOLVER Indicates the solver to be used. 2 3 4
BCS
Boeing Solver (direct solver).
Default for linear static analysis.
MUMPS
MUMPS Solver (direct solver).
Default for nonlinear static and nonlinear transient analysis.
PCG
Preconditioned Conjugate Gradient (iterative solver).
MIXED
Mixed solver using both BCS and PCG.
AUTO
Automatically selects between BCS and PCG.
PARDISO
Activates the PARDISO solver. 10
 
PCON Pre-conditioner type of to be used. 6 11
NO
No Pre-conditioner
DJ
Diagonal Jacobi
FAI (Default for SMP runs)
Factored Approximate Inverse
BLR (Default for DDM runs)
Block Low Rank Appoximation Precondition

(Character)

 
MAXIT Maximum number of iterations.

Default = Number of degrees-of-freedom of the system (Integer > 0 or blank)

 
ITOL Convergence criteria for preconditioned iterative solver.
RROM (Default)
Relative residual of original matrices r < T O L * b MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaauWaaeaaca WHYbaacaGLjWUaayPcSdGaeyipaWJaamivaiaad+eacaWGmbGaaiOk amaafmaabaGaaCOyaaGaayzcSlaawQa7aaaa@425A@ |
RRPM1
Relative residual of preconditioned matrices r < T O L * b MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaauWaaeaaca WHYbaacaGLjWUaayPcSdGaeyipaWJaamivaiaad+eacaWGmbGaaiOk amaafmaabaGaaCOyaaGaayzcSlaawQa7aaaa@425A@ |
RRPM2
Relative residual of preconditioned matrices r < T O L * A * x MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaauWaaeaaca WHYbaacaGLjWUaayPcSdGaeyipaWJaamivaiaad+eacaWGmbGaaiOk amaafmaabaGaaCyqaaGaayzcSlaawQa7aiaacQcadaqbdaqaaiaahI haaiaawMa7caGLkWoaaaa@470F@

If the solver solves A x = b MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaaCyqaiaahI hacqGH9aqpcaWHIbaaaa@39B2@ , the residual is r = A x b MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaaCOCaiabg2 da9iaahgeacaWH4bGaeyOeI0IaaCOyaaaa@3B9A@ .

(Character)

 
TOL Convergence tolerance.

Default = 1.0e-5 for ITOL=RROM, RRPM1 or single precision machine precision (3.0e-8) for ITOL=RRPM2 (Real > 0 or blank)

 
ORDM Ordering method for the MUMPS solver. 8
AMD
Approximate minimum degree method (AMD).
PORD
Provides better performance for a shell-type model.
METIS
METIS package.
SCOTCH
This package is only available on Linux platform.
PTSCOTCH
This package is only available on Linux platform.
AUTO (Default)
Automatically selects the appropriate ordering method.

(Character)

 

Comments

  1. SOLVTYP Bulk Data must be referenced by a SOLVTYP subcase statement. It only applies to static and dynamic subcases.
  2. In optimization of linear static subcases, if iterative solver is selected, and if the responses DRESP1, RTYPE = DISP, LAMA, STRESS, STRAIN, CSTRESS, CSTRAIN, CFAILURE, or FORCE are present the solver is automatically reverted to the direct solver.
  3. MUMPS "Multifrontal Massively Parallel sparse direct Solver" is the default non-symmetric solver for nonlinear static and nonlinear transient analysis (with or without frictional contact); it is also available as an optional symmetric solver for linear static analysis. MUMPS is also supported as an option for nonlinear heat transfer runs. MUMPS is SMP and SPMD parallelized. Generally, MUMPS performance is similar to or better than the performance of BCS, especially for 2D models. The PCG iterative solver is supported for small and large displacement nonlinear analysis.
  4. For an overview of default settings and options for the SOLVER field, refer to Solvers vs Supported Solution Sequences in the User Guide.
  5. The iterative solver is a preconditioned conjugate gradient solver. A Factored Approximate Inverse Preconditioner is the default method. This solver is also SMP parallelized.
  6. The performance of the iterative solver depends on the conditioning of the stiffness matrix. For compact solid models, the iterative solver may perform considerably better than the direct solver in terms of memory usage and elapsed times for a single linear static subcase. In the case of multiple linear static subcases, the iterative solver may perform worse than the direct solver. The breakeven point is at about 4-6 subcases. The performance depends on model, hardware, operating system, and potentially the system load. The performance may be below expectations on Itanium-based computers.
  7. When the automatic solver option (SOLVER=AUTO) has been chosen, PCG is used first, the solver will be changed automatically to direct solver (BCS) if PCG performance is estimated slower than direct solver. In this case, direct solver will be used for the remainder of the run.
  8. For further information about the MUMPS solver ordering method (ORDM) options, refer to the MUMPS 4.10 manual.
  9. The SOLVTYP Bulk and Subcase data pair can be used in a Normal Modes Analysis model to select the direct solver (MUMPS/BCS) used for the solution of the internal linear shift for Lanczos runs. The SOLVTYP Bulk and Subcase data pair can also be used to specify the solver (MUMPS/BCS/PCG) used for Resvec calculation in Modal Frequency Response Analysis subcases. If you wish to select different solvers for the internal linear shift of the Lanczos eigenvalue solution and for the Resvec calculation in the Frequency Response solution, you can split the MFREQ subcase into two different subcases (Normal Modes subcase + Modal FRF subcase) with different SOLVTYP Subcase and Bulk Data pairs in each subcase. The SOLVTYP Bulk and Subcase data pair can also be used in Buckling Analysis and Direct Frequency Response Analysis subcases.
  10. PARDISO may be faster than MUMPS for shell models using SMP with number of threads (nt) greater than 16. PARDISO can be tried as an alternative in models for which MUMPS fails. For static analysis with PARDISO, the constraint elimination method (RIGID entry) is set to LGELIM by default.
  11. The FAI preconditioner is well suited for blocky structures using SMP runs. It is not available for DDM.

    The DJ preconditioner consumes lesser memory, but utilizes larger runtime. It can be used for very large size models to save memory. It is supported for both SMP and DDM.

    The BLR preconditioner balances the attributes of direct solvers and conventional iterative solver (PCG). It provides reasonable performance for ill-conditioned models. It is the default for DDM runs.

  12. This card is represented as a load collector in HyperMesh.