Author Topic: solving unsymmetric tangent  (Read 5511 times)

blackbird

  • Full Member
  • ***
  • Posts: 100
solving unsymmetric tangent
« on: January 14, 2015, 04:27:02 AM »
Dear all.

I implemented an user element with an unsymmetric tangent. While test problems run fine in serial feap, problems show up when calculating in parallel. In serial I used

LOOP,newton, 20
   UTANG,,1     
NEXT,newton

to solve the problem. For parallel I use a "-ksp_type cg -pc_type jacobi" option . Now I read that the conjugate gradient is only solving symmetric problems. So stating UTANG does not have any meaning here? Which solving method would you suggest for unsmmetric tangents?

Thanks in advance

FEAP_Admin

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 993
Re: solving unsymmetric tangent
« Reply #1 on: January 14, 2015, 06:51:04 PM »
You should perform a literature search on iterative methods for unsymmetric problems.  The method of choice will depend on the details of your system of equations.

blackbird

  • Full Member
  • ***
  • Posts: 100
Re: solving unsymmetric tangent
« Reply #2 on: February 02, 2015, 06:43:22 AM »
Ok, lets ask the question another way around:

Is there any difference in the call of UTANG,,1 and TANG,,1 in parallel feap? Are there any additional checks whether the parallel solver in charge is matching with feap's solving commands? Or is the user in charge of checking wether the right setup has been made?

FEAP_Admin

  • Administrator
  • FEAP Guru
  • *****
  • Posts: 993
Re: solving unsymmetric tangent
« Reply #3 on: February 02, 2015, 10:59:04 AM »
In parallel the whole matrix is always assembled so there should be no difference between utan and tang.