Problem that the single-level results differ from the multi-level results

[APTX-4869-core]

Hello AMReX Team!

I'm writing CFD codes in AMReX for generating some points in the flow field, just like what the Particle class does(but I didn't use it).

Now I encountered the problem that when I switched on/off the adaptive regriding, the pressure filed results were different, and the differences were relatively obvious(The velocity fields results were same).

At first, I thought the reason is that I didn't spread the data from the finest level(these points were also defined on the finest level) to the coarser level. But I found the Synchronization in AMReX should have finished the data spreading operation automatically.

So I want to know that if the Particle class has its own data spreading operation, which is different from that of the AMReX Synchronization, to ensure the results in single-level and multi-level are both same?

Thanks for your attention and advice in advance!

Regards,
Arashi

[asalmgren]

AMReX itself doesn’t know anything about pressure per se. Are you using one of the existing CFD codes like IAMR or incflo, or did you write your own? Ann Almgren Senior Scientist; Dept. Head, Applied Mathematics

…

On Thu, Jun 26, 2025 at 12:32 AM HanekawaArashi ***@***.***> wrote: Hello AMReX Team! I'm writing CFD codes in AMReX for generating some points in the flow field, just like what the Particle class does(but I didn't use it). Now I encountered the problem that when I switch on/off the adaptive meshing, the pressure filed results were different, and this difference was relatively obvious(The velocity fields results were same). At first, I thought the reason is that I didn't spread the data from the finest level(these points were also defined on the finest level) to the coarser level. But I found the Synchronization of the AMReX should have finished the data spreading automatically. So I want to know that if the Particle class has its own data spreading operation, which is different from that of the AMReX Synchronization, to ensure the results in single-level and multi-level are both same? Thanks for your attention and advice in advance! Regards, Arashi — Reply to this email directly, view it on GitHub <#4524>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACRE6YSCQWLM4SWFOSOJEDL3FOOX5AVCNFSM6AAAAACAFLRHZ2VHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZYGQ4TMNZYHA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[APTX-4869-core]

Sorry for the incomplete information! I use the IAMR to solve the CFD problems!

[asalmgren]

Can you share which inputs file you used and how you concluded the pressure was different at different levels? It will be easiest to help if we can reproduce what you are seeing Ann Almgren Senior Scientist; Dept. Head, Applied Mathematics

…

On Thu, Jun 26, 2025 at 4:35 AM HanekawaArashi ***@***.***> wrote: Sorry for the incomplete information! I use the IAMR to solve the CFD problems! — Reply to this email directly, view it on GitHub <#4524 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACRE6YWGFPRJOLGXP5422Z33FPLJ3AVCNFSM6AAAAACAFLRHZ2VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGNJYGQ2TOOI> . You are receiving this because you commented.Message ID: ***@***.***>

[APTX-4869-core]

Thanks for your timely reply!

I think it should result from mine lack knowledge of AMReX or my own code error.
So I want to learn from the Particle class to fix the problem, in another word, troubleshouting from the usage of AMReX first.

Actually, I'm trying to simulate the flow past a cylinder problem via the immersed boundary method.
I want to validate it so I tried to calculate the surface friction coefficient Cf from the velocity field, and surface pressure coeffiecient Cp from the pressure field(I used the derived variable avg_pressure to calculate the Cp).
I simulated two cases for simulation time = 2.0s, one with max_level = 2, another one without refinement. But the norm0(I found actually it's infinity norm) got from these two cases were different, as well as the pressure contour:


The first pic is the single-level result and the second one is the multi-level result.
and the basic params excluding those I defined myself:

---

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#            SIMULATION STOP            #
#.......................................#
stop_time                       =  2.0       # Max (simulated) time to evolve
max_step                        =  10000           # Max number of time steps, coarsest level step, 
                                                 # the finer level needs to multiple ref_ratio

ns.stop_when_steady             =  0            # Steady-state solver?
ns.steady_tol                   =  1.e-8

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#           SOLVER SETTINGS             #
#.......................................#
ns.init_shrink                  =  1.0   # Multiply initial dt by this factor

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#         TIME STEP COMPUTATION         #
#.......................................#
ns.fixed_dt                     =  -1.0  # If fixed_dt > 0, use this constant dt
ns.cfl                          =  0.45  # CFL factor

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#            INPUT AND OUTPUT           #
#.......................................#
amr.plot_file                   =  plt
amr.plot_int                    =  100                     # Steps between plot files
amr.plot_per                    =  -1.0                  # Steps between plot files
amr.check_int                   =  20000                  # Steps between checkpoint files
amr.derive_plot_vars            =  mag_vort avg_pressure # Derived quantities to include in plotfiles
amr.restart                     =  ""                    # Checkpoint to restart from

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#               PHYSICS                 #
#.......................................#
ns.gravity		                =  0.0 
ns.vel_visc_coef                =  0.001        # Re = 200
ns.scal_diff_coefs              =  1.0

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#        ADAPTIVE MESH REFINEMENT       #
#.......................................#
amr.n_cell                      =  4096 2048 128  # Grid cells at coarsest AMRlevel
amr.max_level                   =  0           # Max AMR level in hierarchy
amr.regrid_int			        =  10           # Interval (in number of level l timesteps) between regridding
amr.ref_ratio		 	        =  2 2 2 2 2   # Refinement ratio as a function of level
amr.grid_eff                    =  0.90        # Grid efficiency (defaults to 0.70)
amr.n_error_buf                 =  2 2 2 2 2
amr.blocking_factor             =  32          # default = 8
max_grid_size                   =  64          # default = 128 in 2D, and 32 in 3D 
amr.subcycling_mode             =  Optimal        # None, Auto, Optimal, Manual
                                               # usage of Auto will cause very slow computation
                                               # but now it's only suitable for short running (4 steps test)
amr.refinement_indicators_fsi   =  1           # refine on the interface (user-specified area)
amr.refinement_indicators       =  vorticity   # Refinement criterion, use vorticity
amr.vorticity.vorticity_greater =  1.0         # refinement threshold

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#              GEOMETRY                 #
#.......................................#
geometry.prob_lo                =  0.0 0.0 0.0  # Lo corner coordinates
geometry.prob_hi                =  4.0 2.0 2.0  # Hi corner coordinates
geometry.is_periodic            =  0  0  0      # Periodicity x y z (0/1)
geometry.coord_sys              =  0            # 0 for Cartesian, 1 for RZ

# 0 = Interior/Periodic  3 = Symmetry
# 1 = Inflow             4 = SlipWall
# 2 = Outflow            5 = NoSlipWal
ns.lo_bc 	     	            =  1 5 5        # Boundary conditions on the low end of the domain.
ns.hi_bc   	     	            =  2 5 5        # Boundary conditions on the high end of the domain.
xlo.velocity                    =  1.0 0.0 0.0  # inflow velocity

#¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨¨#
#          PROBLEM PARAMETERS           #
#.......................................#
prob.probtype                   =  22005066  # problem for cuibowen
prob.density_ic                 =  1.0       # Constant density initial condition
prob.velocity_ic                =  1. 0. 0.  # Set up a flow, defaults to zero
 

---

So back to the question, I want to refer to the Particle class so that whether it has the extra different data spreading operation from the AMReX Synchronization? Based on this, I can determine if there is a problem with my code itself, or if I need to do additional operation of the data from the finest level to coarser levels.

Any advice is appreciable, thanks sincerely!

Regards,
Arashi

[WeiqunZhang]

What do they look like if you use the same min and max for the color bars in the figures?

[APTX-4869-core]

Thanks for your attention, and sorry I neglected it. From the adjusted contour, it seems that their difference isn't that obvious: