[FLASH-USERS] Growing error in magnetic field when updating fluid variables in wind tunnel simulation

Thu Feb 18 22:31:55 EST 2016

Dear Rukmani,

What kind of Riemann solver are you using?
Can you send me a log file or flash.par?

Thanks,
Dongwook

> On Feb 18, 2016, at 6:59 PM, Rukmani Vijayaraghavan <rukmani at virginia.edu> wrote:
> 
> Hi Jason, Klaus,
> 
> This block-by-block variation is correlated with similar variation in other fluid variables (density, pressure), and this persists even where there is a zero velocity inflow, as well as with a uniform grid, and with both USM and PPM (pure hydro) solvers. Modifying the gravity solver from Multigrid to Multipole doesn't make a difference either. I'm using the FLASH Gamma EOS unit. As far as I've seen, there is no variation in B-field across grid cells adjacent to block / refinement boundaries, this only happens at the inflow edge. 
> 
> I also update the magnetic field face variables (MAG_FACE_VAR and/or MAGI_FACE_VAR), with no effect. Div(B) still seems to be 0. 
> 
> Any other suggestions would be great!
> 
> Thanks,
> Rukmani
> 
> 
> 
>> On 02/18/2016 01:30 PM, Jason Galyardt wrote:
>> Hi Rukmani,
>> 
>> I used a spatially varying wind; the velocity of the wind varies along the boundary, but it has a well-defined, time-independent form. I've also seen problems with more realistic B-field geometries which (to my horror) included step functions in the domain interior. I had to smooth these out to avoid unphysical evolution in those regions. 
>> 
>> I've also seen some modest increase in B-field magnitude for the cells adjacent to a refinement boundary. I haven't reported the latter previously because I haven't had time to figure out what's going on there. You might try setting lrefine_min = lrefine_max to get uniform refinement and see whether that helps (some of our group's simulations do this).
>> 
>> The block by block variation does seem strange. I would expect this kind of variation to be correlated with variation in another variable. How do the other variables look in the problem region?
>> 
>> Another idea: could this variation be tied to the equation of state? If you're using one of the supported FLASH EOS units, you're probably fine.
>> 
>> Regards,
>> Jason
>> 
>> 
>> On Thu, Feb 18, 2016 at 11:07 AM, Rukmani Vijayaraghavan <rukmani at virginia.edu> wrote:
>>> Hi Jason,
>>> 
>>> Thanks! I'm using FLASH 4.2, I'll try using 4.3 to see if that makes a difference. I haven't tried refining on the magnetic variables yet.
>>> 
>>> For the different runtime parameters -- 
>>> 
>>> 1. I've tried cfl = 0.5 and 0.8, but nothing lower yet. I'll check to see if that works.
>>> 
>>> 2. For the Riemann Solver, I've found HLLC to be a bit more dissipative than HLLD, and therefore marginally better at smoothing out the magnetic field at the edges. Ditto with second order MUSCL-Hancock over third order PPM.
>>> 
>>> 3. All the other runtime parameters are mostly the same. I don't refine on the magnetic variables, but I tried higher overall lrefine_min (to make sure the outer edges get further refined) and it didn't help -- the same block-based discontinuity persists.
>>> 
>>> 4. I'm using a constant wind inflow for this particular run. One thing I checked to see was if there was a round off error in reading my input variables into double precision arrays, and this tiny "seed" instability might grow, but it doesn't seem to be an issue. What is strange is that the value (and sign) of the initial instability varies block-by-block. In your simulations, did you use a constant wind?
>>> 
>>> Thanks,
>>> Rukmani
>>> 
>>> 
>>>> On 02/18/2016 09:38 AM, Jason Galyardt wrote:
>>>> Hi Rukmani,
>>>> 
>>>> I've had some similar issues with MHD runs. You didn't mention which version of FLASH you're using, but I've found the latest (v4.3) to be a bit more stable than v4.2 or v2.5. As for runtime parameters, found the following combination to be helpful:
>>>> 
>>>> #~~~~
>>>> # Refine on the magnetic variables:
>>>> refine_var_1 = "dens"
>>>> refine_var_2    = "magp"
>>>> # -OR-
>>>> # refine_var_2 = "magx"
>>>> # refine_var_3 = "magy"
>>>> # refine_var_4 = "magz"
>>>> # prefer higher refinement, according to magp (default refine_cutoff_X = 0.8)
>>>> refine_cutoff_2 = 0.7
>>>> # refine_cutoff_3 = 0.7
>>>> # refine_cutoff_4 = 0.7
>>>> 
>>>> # Lower CFL: between 0.25 and 0.5
>>>> cfl = 0.5
>>>> 
>>>> # Use second order MUSCL-Hancock reconstruction scheme
>>>> order = 2
>>>> 
>>>> # I've mostly used the "hybrid" slope limiter, but occasionally I've found the "minmod" useful in particularly difficult situations 
>>>> slopeLimiter    = "hybrid"
>>>> 
>>>> # use flattening (dissipative) (originally for PPM)
>>>> use_flattening    = .true. 
>>>> 
>>>> # Use high order algorithm for E-field construction
>>>> E_modification  = .true.
>>>> 
>>>> # Update magnetic energy using staggered B-fields
>>>> energyFix       = .true.
>>>> 
>>>> # Prolongation method (injecton_prol, balsara_prol) -- Using Balsara's method is particularly critical, in my experience.
>>>> prolMethod      = "BALSARA_PROL"
>>>> 
>>>> # For the Riemann solver, I use HLLD for MHD runs, and HLLC for pure hydro runs.
>>>> RiemannSolver    = "HLLD"
>>>> #~~~~
>>>> 
>>>> What sort of inflow conditions have you implemented? Small non-linearities in the                             inflow can grow into large unphysical features over time (I've seen this happen in my own simulations). So, it's worth checking your boundary condition code for undesirable features. In any case, I hope this helps.
>>>> 
>>>> Sean: is the E_upwind option available for the unsplit MHD solver in FLASH 4.3? My recollection is that it caused some problems in previous versions....
>>>> 
>>>> Regards,
>>>> Jason
>>>> 
>>>> 
>>>>> On Wed, Feb 17, 2016 at 9:22 PM, Rukmani Vijayaraghavan <rukmani at virginia.edu> wrote:
>>>>> Hi everyone,
>>>>> 
>>>>> I've come across an error when updating fluid variables at the inflow edge of a wind tunnel simulation. I'm running a simulation of a galaxy (with active dark matter particles, gas, and passive particles) in a box, whose fluid is initialized to be identical to the incoming wind (with vx, vy, vz = 600 km/s, 0, 0). There is a small error (on the order of 1%) when updating grid cells near the inflow boundary (with both USM and PPM solvers), and this error is spatially correlated with  block boundaries. While this error itself is tolerable as far as the density and pressure go, this has bad consequences for the magnetic field which grows as the wind propagates through the box (see attached figure, xl_boundary). This figure shows slices of Bx at two timesteps (annotated with block boundaries and magnetic field vectors). The dynamic range of Bx in this image has been reduced to highlight these discontinuities. At the timesteps shown in the attached image, the fluctuations in Bx are ~1%, but grow with time up to order unity. I've tried a variety of Riemann solvers (HLLC, HLLD, Roe, Hybrid), slope limiters (mc, minmod, etc.), interpolation orders, prolongation methods, turning on and off specific USM switches, but nothing seems to solve this issue so far. Has anybody else dealt with and/or successfully solved this issue?
>>>>> 
>>>>> Thanks,
>>>>> Rukmani
>>>>> 
>>>>> -- 
>>>>> Rukmani Vijayaraghavan
>>>>> NSF Astronomy & Astrophysics Postdoctoral Fellow
>>>>> University of Virginia
>>>>> rukmani at virginia.edu
>>> 
>>> -- 
>>> Rukmani Vijayaraghavan
>>> NSF Astronomy & Astrophysics Postdoctoral Fellow
>>> University of Virginia
>>> rukmani at virginia.edu
> 
> -- 
> Rukmani Vijayaraghavan
> NSF Astronomy & Astrophysics Postdoctoral Fellow
> University of Virginia
> rukmani at virginia.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20160218/f4587707/attachment-0001.htm>