Reply To: When a consistency check reveals a difference in results, which result is true?

[Pedram Samadian]

Hello,

Thank you for contacting us. The difference between the predictions is related to the randomness of numerical round-off in finite element simulations, which refers to small, non-deterministic variations introduced when representing real numbers in a computer.

Computers have finite precision and thus cannot represent every possible decimal value exactly. Instead, they use an approximation, which leads to small round-off errors. In complex simulations, these round-off errors can sometimes create seemingly random effects, particularly in highly sensitive computations.

In the LS-DYNA software, such an issue can be raised when running simulations in the shared memory parallel mode. To address it, the use of the *CONTROL_PARALLEL card with the consistency option on is recommended. In any given problem with the consistency option off (i.e. CONIST = 2) in the *CONTROL_PARALLEL card, summations in calculations occur in a different order from run to run, so the round-off becomes random. However, with the included consistency flag (i.e. CONST = 1), all contributions to global vectors are enforced to be summed in a precise order, resulting in identical results (or nearly so) regardless of the number of processors used.

Further information about the consistency checking in LS-DYNA can be found in the remark section of the *CONTROL_PARALLEL card in the LS-DYNA User Manual Vol I (URL: LSDYNA Manuals).

Based on the provided information, the predictions with the consistency check should be more accurate; however, comparison with the experimental results or analytical calculations is recommended. 

Thank you,
Pedram