Real Eigenvalue Extraction Data
Description: Defines data needed to perform real eigenvalue (vibration or buckling) analysis.
Format:
Example:
| Field | Definition | Type | Default | ||||||
|---|---|---|---|---|---|---|---|---|---|
| SID | Set identification number. | Integer > 0 | Required | ||||||
| V1, V2 | For vibration analysis: frequency range of interest. For buckling analysis: eigenvalue range of interest. | Real or blank, V1 < V2 | See Remark 5 | ||||||
| ND | Number of roots desired. | Integer > 0 | See Remark 5 | ||||||
| SCHECK | Sturm sequence check, one of the following character variables: YES, NO, or AUTO. See Remark 7. | Character | AUTO | ||||||
| NIVEC | Number of iteration vectors. See Remark 8. | Integer > 0 | See Remark 8 | ||||||
| SHFSCL | Estimate of the first flexible mode natural frequency. See Remark 9. | Real or blank | See Remark 9 | ||||||
| NORM | Method for normalizing eigenvectors, one of the following character variables: MASS or MAX
|
Character |
|
||||||
| MAXITER | Maximum number of iterations. See Remark 10. | Integer ≥ 0 | 0 | ||||||
| CTOL | Eigenvalue convergence tolerance. | Real or blank | See Remark 11 | ||||||
| ADDITER | Number of additional iterations after convergence. See Remark 12. | Integer ≥ 0 | 1 | ||||||
| ADDIVCV | Number of additional iteration vectors past the number of roots desired or the included range of interest that must also converge. See Remark 12. | Integer ≥ 0 | 5 | ||||||
| CTRLOPT | Controls solver specific operations during eigenvalue extraction. See Remark 13. | 1 ≤ Integer ≤ 4 | See Remark 13 | ||||||
| ORTOPT | Option for full or partial mass re-orthogonalization after each Lanczos iteration, one of the following character variables: FULL, PARTIAL, or AUTO. See Remark 14. | Character | AUTO |
Remarks:
- Real eigenvalue extraction data sets must be selected with the Case Control command METHOD = SID.
- The units of V1 and V2 are cycles per unit time in vibration analysis, and are eigenvalues in buckling analysis. In buckling, each eigenvalue is the factor by which the prebuckling state of stress is multiplied to produce buckling in the shape defined by the corresponding eigenvector.
- NORM = MASS is ignored in buckling analysis and NORM = MAX will be applied.
- Eigenvalues are sorted on order of magnitude for output. An eigenvector is found for each eigenvalue.
- In vibration analysis, if V1 < 0.0, the negative eigenvalue range will be searched. (Eigenvalues are proportional to Vi squared; therefore, the negative sign would be lost.) This is a means for diagnosing improbable models. In buckling analysis, negative V1 and/or V2 require no special logic.
- The roots are found simultaneously and sorted in increasing order for each subspace or Lanczos iteration. The number and type of roots to be found can be determined from the following table.
V1 V2 ND Number and Type of Roots Found V1 V2 ND Lowest ND roots or all in range, whichever is smaller V1 V2 blank All in range V1 blank ND Lowest ND roots in range [V1, + ∞ ] V1 blank blank Lowest root in range [V1, + ∞ ] blank blank ND Lowest ND roots in range [- ∞ , + ∞ ] blank blank blank Lowest root blank V2 ND Lowest ND roots below V2 blank V2 blank All below V2 - SCHECK controls whether a Sturm sequence check is performed. The Sturm sequence check determines if any roots were missed during eigenvalue extraction. Setting SCHECK equal to 0 or NO skips the Sturm sequence check and avoids an additional stiffness matrix factorization thus reducing analysis time. Setting SCHECK equal to 1 or YES performs the check and will output a warning message if any modes were missed. The default setting of AUTO will always perform the check when the subspace eigensolver is selected and only for models smaller than EXTRACTAUTOSIZE when the Lanczos eigensolver is selected. (See Section 2, Initialization, for more information on EXTRACTAUTOSIZE.)
- When the subspace eigensolver is selected, NIVEC specifies the number of additional iteration vectors and is defaulted to 12. Increasing this value may result in a lower number of subspace iterations required but will require more memory and more solves per subspace iteration. When the Lanczos eigensolver is selected, this option controls the Lanczos block size and the default is determined automatically. A value of 9 or 12 may increase performance for models where a large number of modes will be extracted. The maximum value for the Lanczos eigensolver is 120.
- Specifying SHFSCL = 0.0 may improve accuracy and performance. If this field is blank, a non-zero value for SHFSCL is estimated automatically to handle unconstrained or poorly constrained structures in vibration analysis.
- MAXITER is used to limit the number of subspace or Lanczos iterations to be performed. The default zero setting forces the eigensolver to iterate until convergence is reached.
- The CTOL default is dependent on the
OPTIMIZESETTINGS directive setting. The following table gives the various values. The default for OPTIMIZESETTINGS is NONE.
OPTIMIZESETTINGS Value CTOL Value SPEED 1.0E-5 ACCURACY 1.0E-7 BOTH 1.0E-6 NONE 1.0E-6 - ADDITER and ADDIVCV are used to prevent missing roots. ADDITER defines the number of additional iterations that will be forced even after all roots desired have converged. ADDIVCV defines how many roots past the desired number or range of interest must converge. A value greater than 1 is recommended when roots are closely spaced. Larger values may result in additional subspace iterations.
- CTRLOPT controls where the Lanczos eigensolver intermediate results are stored (in memory or on disk) and what solver mode is used (iterative or direct). Higher settings require more memory but may increase performance significantly. The default setting is the eigensolver selects the best method based on available memory. If the SPARSEITERMETHOD model parameter is set to DIRECT, the default will be a CTRLOPT setting of 4. If set to ITERATIVE and the model consists of mostly parabolic tetrahedron elements, the default will be a setting of 1. (See Section 5,
Parameters, for more information on
SPARSEITERMETHOD.) The following table gives the various options.
CTRLOPT Setting Intermediate File Storage Location Solver Mode 1 Disk Iterative 2 Memory Iterative 3 Disk Direct 4 Memory Direct - ORTOPT controls whether a full or partial mass re-orthogonalization is performed after each Lanczos iteration. Partial re-orthogonalization increases performance for models where a large number of modes (greater than 100) are requested. Partial re-orthogonalization, however, may result in a small degradation in accuracy. The AUTO setting will use partial re-orthogonalization when residual vectors are requested via the RESVEC model parameter or for models larger than EXTRACTAUTOSIZE when either an eigenvalue range is specified or the number of modes requested is greater than 100. (See Section 2, Initialization, for more information on EXTRACTAUTOSIZE and Section 5, Parameters, for more information on RESVEC.)