Learning Forum

[John Farrell]

Hi, 

Transient simulation a rotating container with fluid inside.

The radial velocity is constant for a period of time after which I am attempting to simulate a near step change in the radial velocity (0.25s decceleration to 0 [rad/s]). 

To decrease simulation time I want to

• Have a larger timestep and less max iterations per timestep for a period before the change in radial velocity
• Decrease the timestep size and increase the max iterations per timestep during the decelleration period
• Return to the original timestep and max iterations per timestep for the remainder of the simulation. 

I created an expression that follows the desired radial velocity profile using conditional statements.

My question is, if I were to stop the simulation before the decelleration period, restart with a much lower timestep value and increased iterations per timestep will this affect my simulations convergence? and similarly after the decelleration period, will the abrupt increase in timestep size again affect convergence? 

 

Thanks in advance,

[Murari Iyengar]

Hi John, increasing iterations per timestep will only help the solution converge better, if it isn't converging in the specified number of iterations already. As for decreasing and increasing timesteps, you also need to factor in the CFL number before doing so. CFL~1 is ideal so based on this, you can change your timestep size. 

[John Farrell]

Hi again, 

CFL at my current timestep size is < 1 throughout the simulation beginning and end. 

During the decelleration period I decreased the timestep size to avoid convergence issues resulting in CFL << 1, I am seeing large ocsillations in scaled residuals plot from the beginning of each timestep to the end (~1e-3 to ~1e-13 continuity, ~1e-5 to ~1e-18 y-velocity, and ~1e-8 to ~1e-18 x & z-velocity), is this indicative of a setup problem or should I only be concerned with the final value reached at the end of each timestep? The final iteration of each timestep reaches the same value as the previous ones.

Also, are there negatives to prescribing a timestep that results in CFL << 1 for simulations?

Thanks in advance,

(Sorry for the spam, I replied to thread as opposed to your comment specifically)

 

[John Farrell]

 

 

[Murari Iyengar]

That is normal for transient cases. The residuals will shoot up at the start of a new timestep and decrease as the iterations carry on. You need to look at the final value of the residual at the end of a timestep to see if the solution is converging. 1e-13 is very good convergence for continuity indicating a fully converged flow solution. The downside to having CFL<<1 is that your simulation will take longer to complete/use more resources than required to reach the intended flow time.