New Work

1. Analysis of runs 132 (high res), 133 (like 125 of last post but max refinement volume that decreases with time), 135 (like 133 but no accretion).
2. New run 136 to test importance of RG damping (in progress).

Summary of New Results

1. Each of the runs analyzed gives useful information:
   • Run 132 shows that doubling the resolution matters, e.g. increases frequency of oscillations, decreases final separation.
   • Run 133 shows that it is quite reasonable to use max refinement within a radius min(5e12cm, 1.5*particle_separation) about the primary.
   • Run 135 shows that accretion has a small effect on particle separation (decreases) and orbital frequency (increases).
2. New run 136: first 4 days so far consistent with runs that start from "relaxed" RG.

Analysis of runs 133, 135, 132

Old run 088 (for comparison)

Binary run 088 with longer simulation time and lower resolution in ambient medium
Relaxation run: 062
First frame: 75 (5 RG freefall times, when velocity damping ended)
Last frame: 161
Total simulation time: 20 days
Machine and partition: Comet normal
Number of cores: 1728 (2 cores per task to increase memory)
Total wall time: 4 days
Hydro BCs: extrapolated
Poisson BCs: multipole expansion
Box size: L=4e13 cm (575 Rsun)
Base resolution: 9.0 Rsun (64³ cells)
Highest resolution: 0.29 Rsun (2048³ cells, 5 levels AMR)
AMR implementation: set internally by astrobear
Max resolution zone: n/a
Buffer zones: n/a
Softening length: 4.8 Rsun
Ambient density: 6.7e-9 g/cc
Ambient pressure: 10⁶ dyne/cm²

Old run 125 (for comparison)

Binary run 125 with longer simulation time and lower resolution in ambient medium
Relaxation run: 096
First frame: 75 (5 RG freefall times, when velocity damping ended)
Last frame: 335
Total simulation time: 5.2e6 s or 60 days ~6.5 RG sound-crossing times (260 frames)
Machine and partition: Bluehive standard
Number of cores: 120
Total wall time: about 8.5 days
Hydro BCs: extrapolated
Poisson BCs: multipole expansion
Box size: L=4e13 cm (575 Rsun)
Base resolution: 9.0 Rsun (64³ cells)
Highest resolution: 0.29 Rsun (2048³ cells, 5 levels AMR)
AMR implementation: set by hand to have max level around point particles
Max resolution zone: within 50 Rsun of primary and within a cylinder of radius 50 Rsun and height 50 Rsun around secondary
Buffer zones: 0 cells (no buffer zones)
Softening length: 2.4 Rsun
Ambient density: 6.7e-9 g/cc
Ambient pressure: 10⁶ dyne/cm²

Run 133 to test making refinement zone decrease with time:

Binary run 133 similar to run 125 but now the refinement zone changes with time
Relaxation run: 096
First frame: 75 (5 RG freefall times, when velocity damping ended)
Last frame:
Total simulation time: 84 days (up to frame 439)
Machine and partition: Bluestreak standard
Number of cores: 8192 (2 cores per task to increase memory)
Total wall time: 6 days
Hydro BCs: extrapolated
Poisson BCs: multipole expansion
Box size: L=4e13 cm (575 Rsun)
Base resolution: 9.0 Rsun (64³ cells)
Highest resolution: 0.29 Rsun (2048³ cells, 5 levels AMR)
AMR implementation: set by hand to have max level around point particles
Max resolution zone: within min(5e12cm, 1.5*particle_separation)
Buffer zones: 2 cells
Softening length: 2.4 Rsun
Ambient density: 6.7e-9 g/cc
Ambient pressure: 10⁶ dyne/cm²
DefaultAccretionRoutine=2 (Krumholz)

Comparison between runs 125(top left), 133(top right), O+16a(bottom)

Comments:

• The runs are very similar but there is a slightly smaller final separation for run 125(less aggressive refinement)
• This suggests that our refinement criteria for run 133 was probably slightly too aggressive.
• However, the otherwise close agreement tells us that decreasing the refinement zone with time to be a sphere centred on p1 with radius ~2 times particle separation is reasonable.

Run 135 to test case where no accretion onto secondary is permitted:

Binary run 135 similar to run 133 but now DefaultAccretionRoutine=0 instead of 2. Also suppress generation of new sink particles.
Relaxation run: 096
First frame: 75 (5 RG freefall times, when velocity damping ended)
Last frame: 900
Total simulation time: 191 days
Machine and partition: bluehive2.5 standard
Number of cores: 120
Total wall time: 2.8 days
Hydro BCs: extrapolated
Poisson BCs: multipole expansion
Box size: L=4e13 cm (575 Rsun)
Base resolution: 9.0 Rsun (64³ cells)
Highest resolution: 0.29 Rsun (2048³ cells, 5 levels AMR)
AMR implementation: set by hand to have max level around point particles
Max resolution zone: within min(5e12cm, 1.5*particle_separation)
Buffer zones: 2 cells
Softening length: 2.4 Rsun
Ambient density: 6.7e-9 g/cc
Ambient pressure: 10⁶ dyne/cm²
Default Accretion Routine=0 (No accretion)

Comparison between runs 133(top left), 135(top right), O+16a(bottom)

Comments:

• The results of 135 and 133 are similar, except:
  • Run 135 (no accretion) has slightly smaller separation at a given time.
  • Run 135 (no accretion) has slightly larger frequency of oscillations.
• Thus accretion cannot fully explain the discrepancy with the O+16a results, but removing accretion does make results slighly closer to those of O+16a, who did not have accretion.
• We must think more about accretion, but it is not making a huge difference at present.

Run 132 that uses twice as high resolution and 10x lower ambient pressure as previous runs:

Binary run 132 with double max resolution, lower resolution in ambient medium, 10x smaller ambient pressure than run 116
Relaxation run: 129
First frame: 75 (5 RG freefall times, when velocity damping ended)
Last frame:
Total simulation time:
Machine and partition: Stampde 1 normal (running, completed up to frame 215, or 32 sim-days)
Number of cores: 1024
Total wall time: 8 days (starting from frame 75 of relaxation run)
Hydro BCs: extrapolated
Poisson BCs: multipole expansion
Box size: L=4e13 cm (575 Rsun)
Base resolution: 2.25 Rsun (256³ cells)
Highest resolution: 0.14 Rsun (4096³ cells, 4 levels AMR)
AMR implementation: set by hand to have max level around point particles
Max resolution zone: within 5e12 cm (71.87 Rsun) of primary center and within a cylinder of radius 20 Rsun and height 20 Rsun around secondary center. After t~11d (~frame 123) refinement radius around primary was halved to 2.5e12cm (36Rsun). After t~31d (frame 210) halved again to 1.25e12cm (18Rsun).
Buffer zones: 2 cells
Softening length: 2.4 Rsun
Ambient density: 6.7e-9 g/cc
Ambient pressure: 10⁵ dyne/cm²
DefaultAccretionRoutine=2 (Krumholz)

NOTE: For the last 4 days of wall time the code ran very slow, probably because it was creating many many low mass particles due to the Jeans criterion. I have now commented out this part of the code, so it does not create particles. It is pending on stampede from chombo file 208, before extra particles were created.

Comparison of run 132 with Ohlmann+16a run:
​Table

Comments:

• The softening length of the point particles decreases with time in the O+16a simulation.
  • It is not clear whether this procedure is justified.
  • The main reason to improve the resolution with time may be to resolve the decreasing softening length.
  • Therefore, if we keep the softening length constant, improving the resolution with time may not be necessary or productive.
• O+16a initially resolves the softening length by 20 cells, whereas we resolve it by 17 cells. They advocate >10 cells.
• At 120 days, O+16a resolves the softening radius by 35 cells.
• It may be worth doing a run with slightly larger or smaller resolution to ensure that we are adequately resolving the softening length.

Comparison between runs 088(top left), 125(top right), 132(bottom left), O+16a(bottom right)

Comments:

• The 1st minimum, 2nd maximum, and 2nd minimum are located roughly at separations, with softening length denoted as h:
  • Run 088 (res=0.29Rsun, h=4.8Rsun, cells/h=17): 14 , 21 , 12 Rsun
  • Run 125 (res=0.29Rsun, h=2.4Rsun, cells/h= 8): 13.5, 18 , 14.5 Rsun
  • Run 132 (res=0.14Rsun, h=2.4Rsun, cells/h=17): 14 , 18.5, 11 Rsun
  • O+16a (see above table) : 10 , 23.5, 11 Rsun
• These are found roughly at times:
  • Run 088 (res=0.29Rsun, h=4.8Rsun, cells/h=17): 12.5, 15.5, 18 days
  • Run 125 (res=0.29Rsun, h=2.4Rsun, cells/h= 8): 12.5, 14.5, 17 days
  • Run 132 (res=0.14Rsun, h=2.4Rsun, cells/h=17): 12.5, 14.5, 17 days
  • O+16a (see above table) : 13 , 16 , 19 days
• From the difference between 125 and 132, we conclude that h is not adequately resolved in run 125 (8.5 cells/h vs 17 cells/h)
• O+16a advocates >10 cells/h, so this is consistent with their result.
• In our case, we still cannot tell if 17 cells/h is sufficient.
• We could run 088 with double the resolution to see if it converges.
• I will also continue to run 132 but it will be slow as the queue time in stampede 1 has increased as nodes are being transferred to stampede 2.

Run 136 to test how much relaxation makes a difference (IN PRORESS):

Binary run 136 similar to run 135 but now start from frame 0 of relaxation run 096 instead of frame 75 (so no damping).
Relaxation run: 096 (but no relaxation! Just start with initial profile
First frame: 0
Last frame: 18, running
Total simulation time: 4 days
Machine and partition: bluehive2.5 standard
Number of cores: 120
Total wall time: 21 hours, running
Hydro BCs: extrapolated
Poisson BCs: multipole expansion
Box size: L=4e13 cm (575 Rsun)
Base resolution: 9.0 Rsun (64³ cells)
Highest resolution: 0.29 Rsun (2048³ cells, 5 levels AMR)
AMR implementation: set by hand to have max level around point particles
Max resolution zone: within min(5e12cm, 1.5*particle_separation)
Buffer zones: 2 cells
Softening length: 2.4 Rsun
Ambient density: 6.7e-9 g/cc
Ambient pressure: 10⁶ dyne/cm²
Default Accretion Routine=0 (No accretion)

Discussion

• Possible reasons for discrepancy with O+16a in (my guessed) order of importance
  1. resolution (especially number of cells per softening length h) (a new run will test this).
  2. softening length.
  3. different RG relaxation runs with different type codes.
  4. initial spin of RG (a new run will test this).
  5. very different ambient densities (I have an old run that we can use to check this).
  6. different accretion algorithms.
  7. small differences in particle masses.
  8. some reflection off of boundaries at late times in our smaller box.
• Probably run 132 is at the highest resolution we're going to get for the paper (especially if we need to do multiple runs).
• But we need to make sure that the softening length h is adequately resolved by checking that results are converged.

Next steps

• Continue to run high res run 132 (with particle creation turned off!).
• Continue to run 136 (no relaxation run).
• Make separation plot for old run 120 to see if ambient density matters.
• New run with initial spin of RG at 95% Keplerian to test what difference this makes.
• New run like run 088 but with twice as high resolution to test the convergence with cells/h.
• Continue to improve and extend the outputting of relevant data to various files.
• Clean up code and divide into different modules for relaxation runs and binary runs.
• Think more about the accretion routine, Krumholz vs. Federrath vs. no accretion….