Tests 8/31
Fixed the temps in at least one of these. Frames for all. VisIt didn't behave as I expected, so missing the first half of 2x1011, 2x1012.
2x1011
2x1011, no Lyman
2x1012
2x1012, no Lyman
2x1013
2x1013, no Lyman
2x1013, X=0
Issues 8/30
Simulations to Run
- Photon fluxes of 2x1011, 2x1012, 2x1013 (maximum), with and without Lyman
- X = 0 inside planet, 2x1013 flux with Lyman
Original Results
Line transfer routines crashing was a problem with density and pressure protections not recalculating ionization state. Fixed (see ProtectionOptions).
Longer run doesn't look any better, though. Note that the temperature may not be exact (may be off by up to a factor of 2).
After 4 days
AstroBEAR/ATHENA debugging
Initial Conditions
Initial conditions for the planet are finally correct. The ambient media still differ - the planet ends slightly earlier, and the profile is very steep near the edge.
Density
Temperature
Pressure
Ionization Fraction
Radiative Transfer
- The optical depths are close, but not quite identical. At ~300 s,
Working backwards, I find \sigmaH = ~6.28x10-18 for AstroBEAR, and ~6.14x10-14 for ATHENA - tested, and the difference isn't significant for the optical depth:
(ATHENA in blue, AstroBEAR in orange)
- Ionization/recombination rates and heating/cooling rates are qualitatively similar, but still differ by 1-2 orders of magnitude.
Ionization
(Without advective terms)
Theory Comparisons
It appears that the actual data from ATHENA doesn't match the theoretical calculation (using the data for n, X, and \tau in the equations on the LineTransfer page).
Heating
(Without advective terms)
- Electron densities are correct:
Dynamics
For the first time, the dynamics are beginning to look sort of similar. Radial velocities are too large outside of the planet for AstroBEAR:
And the evolution of the temperature on the day side is correct:
Update on CE simulations
New Work
- I performed a new binary run with half the softening length but other parameters the same (run 116 with relaxation run 096) except that I also changed the initial positions and velocities of the particles so that the frame of reference is effectively rotated by 180 degrees.
Summary of New Results
- For the binary runs, the code slows down after about t = 10-15 days and the chombo files get much bigger, probably due to increasing refinement.
- The separation vs time graph is similar to the old run, and about equally similar to O+16a.
Detailed Results
Binary run 96/116 with half softening length of old run 062/088
Damp116) Extrapolated hydro BCs, Multipole expansion Poisson BCs, ambient dyne/cm , ambient density g/cc.
(Stampede 1 normal 1024 cores)
( cm, , 5 levels AMR)
2d density slice zoomed
2d density slice with mesh
- How does halving the softening length change the results? Let us compare the old run (062/088) with the new run (096/116) with softening length 2.4Rsun instead of 4.8Rsun (for both primary and secondary).
2d density comparison of 062/088 (left) and new run 096/116 (right)
Below are snapshots of density in the orbital plane from t = 10 days (left) and t = 20 days (right). Top row: old simulation 062/088. Next row: new simulation 096/116 with half the softening length. Bottom fig 1 from O+16a. In the O+16a figure, the '+' denotes the primary core particle, while the 'x' denotes the secondary point particle. The core and secondary are denoted as '0' and '1' respectively in my plots.
We see that the two runs are qualitatively very similar, as expected. How do they compare in detail?
On the bottom below I've plotted Fig 1 of Ohlmann+16a and the equivalent figure with these simulations for comparison. Note that the new run went up to 21 days while the old run went up to 20 days. In my plot of separation, I've shown the softening radius as a dotted horizontal line. Circles represent the softening radius, while the green dot shows the initial center of mass of the two stars. The very small green square in the upper right shows the smallest resolution element, which is the same for both runs (about 0.29 Rsun).
Discussion
- Up to a few days after the first minimum, the results look almost identical. Softening length begins to play a role thereafter.
- However it is not clear which run is more accurate because the softening length is not as well-resolved in the new run.
- The new run (096/116) takes about 6 days of wall time (including 2 days for the relaxation run).
- The code becomes even slower after t = 10 days of the binary run, probably because the AMR is refining to the max level over a larger and larger volume.
- Regardless, we need to get higher resolution, larger box size, longer relaxation runs, and longer binary simulation time.
- This will require that we force the resolution to be much lower away from the point particles.
Conclusions
- Halving the softening length produces obvious quantitative differences, but only starting toward the end of the first orbit.
- However, it is unclear whether the differences between the two runs with different softening lengths are due to differences in the softening length per se or to differences in the ratio of softening length to resolution.
Next Steps
- I managed to perform a low-ambient density run (rho = 1E-10 g/cm3 rather than 6.7E-9 g/cm3) on stampede 1 while on vacation…it will take a few more days to complete. This will tell us what difference the ambient density makes.
- We need to modify the code to make it able to refine to a high level only around the (moving) point particles.
- Then it will be feasible to increase the max resolution and isolate the effects of resolution and softening length.
Useful Papers on Common Envelope Evolution
Note: linked pdf files may contain my text highlighting.
Ricker+Taam
Ricker & Taam 2008, ApJ 672:L41, The interaction of stellar objects within a common envelope
Ricker & Taam 2012, ApJ 746:74, An AMR study of the common-envelope phase of binary evolution
De Marco++
Passy et al. 2012, ApJ 744:52, Simulating the common envelope phase of a red giant using smoothed-particle hydrodynamics and uniform-grid codes
Staff et al. 2016a, MNRAS 455, 3511, Hydrodynamic simulations of the interaction between an AGB star and a main-sequence companion in eccentric orbits
Staff et al. 2016b, MNRAS 458, 832, Hydrodynamic simulations of the interaction between giant stars and planets
Kuruwita et al. 2016, MNRAS 461, 486, Considerations on the role of fall-back discs in the final stages of the common envelope binary interaction
Iaconi et al. 2017a, MNRAS 464, 4028, The effect of a wider initial separation on common envelope binary interaction simulations
Galaviz et al. 2017, ApJS 229:36, Common envelope light curves. I. Grid-code module calibration
Iaconi et al. 2017b, arXiv:1706.09786v1, The effect of binding energy and resolution in numerical simulations of the common envelope binary interaction
Ohlmann++
Ohlmann et al. 2016a, ApJ 816:L9, Hydrodynamic moving-mesh simulations of the common envelope phase in binary stellar systems
Ohlmann et al. 2016b, MNRAS 462:L121, Magnetic field amplification during the common envelope phase
Ohlmann et al. 2017, A&A 599, A5, Constructing stable 3D hydrodynamical models of giant stars
MacLeod++
MacLeod & Ramirez-Ruiz 2015a, ApJ 798:L19, On the accretion-fed growth of neutron stars during common envelope
MacLeod & Ramirez-Ruiz 2015b, ApJ 803:41, Asymmetric accretion flows within a common envelope
MacLeod et al. 2017a, ApJ 835:282, Lessons from the Onset of a Common Envelope Episode: the Remarkable M31 2015 Luminous Red Nova Outburst
MacLeod et al. 2017b, ApJ 838:56, Common envelope wind tunnel: coefficients of drag and accretion in a simplified context for studying flows around objects embedded within stellar envelopes
Murguia-Berthier et al. 2017, arXiv:1705:04698v1, Accretion disk assembly during common envelope evolution: implications for feedback and LIGO binary black hole formation
Ivanova++
Ivanova et al. 2013a, A&ARv 21, 59 Common envelope evolution: where we stand and how we can move forward
Ivanova et al. 2013b, Science 339, 433 Identification of the long-sought common-envelope events
Nandez et al. 2014, ApJ 786, 39, V1309 Sco—Understanding a merger
Ivanova et al. 2015, MNRAS 447, 2181 On the role of recombination in common envelope ejections
Nandez et al. 2015, MNRAS 450, L39, Recombination energy in double white dwarf formation
Nandez & Ivanova 2016, MNRAS 460, 3992 Common envelope events with low-mass giants: understanding the energy budget
Ivanova & Nandez 2016, MNRAS 462, 362 Common envelope events with low-mass giants: understanding the transition to the slow spiral-in
Update on common envelope simulations
Recap of Last Post
- I had performed a run up to a simulation time of 20 days which had the same parameters as the Ohlmann+16a simulation, except that:
- the softening length is larger by a factor of 2
- the smallest resolution element is larger by a factor
- the RG primary is not rotating initially
- the ambient density is equal to that at the surface of the RG, which is orders of magnitude larger than in Ohlmann+16a
- the binary run starts just after velocity damping has completely turned off (rather than waiting 5 more dynamical times) and the dynamical time used is the free-fall time, which is about 3/8 of the sound-crossing time used by them
- In the binary run that we had done, the closest separation was already down to about 2.5 smoothing radii. In Ohlmann+16a they set a floor that the softening radius should not exceed 1/5 of the separation. Their softening radius thus changed dynamically.
New Work
- I attended the "Evolved Stars: Role of Binarity" conference in Nice and presented this work.
- I did a relaxation run for the RG star with the softening length reduced by a factor of two but the resolution the same.
- I tried several runs with the same parameters as for the original relaxation run but with lower ambient density.
Summary of New Results
- The talk went well. The slides are available here.
- With limited computing resources, it doesn't make much sense to try one change at a time, so I worked on the density issue (using bluehive).
- The lower ambient density runs are very slow. I present the results of tests below.
- In parallel, Baowei, is trying to optimize the code on Comet.
Detailed Results
- Run with half of previous softening length (2.4Rsun instead of 4.8Rsun) but same resolution 0.29 Rsun: central density decreases by 35% from to s when damping is turned off. This can be compared with 11% for the original run with 4.8Rsun softening length. Should resolution be doubled? For comparison Ohlmann+16a used softening radius of 2.8Rsun and resolution 0.14Rsun for relaxation run.
Relaxation run with single RG star
Damp104) Extrapolated hydro BCs, Multipole expansion Poisson BCs, ambient dyne/cm , ambient density g/cc.
(bluehive standard 120 cores)
( cm, , 5 levels AMR)
lTrackDensityProtections=T, lTrackPressureProtections=T, MinDensity=1d-14
2d density + velocity
2d pressure
2d temperature
Damp105) Extrapolated hydro BCs, Multipole expansion Poisson BCs, ambient
(bluehive standard 120 cores)
( cm, , 5 levels AMR)
2d density + velocity
2d pressure
2d temperature
Discussion
- The run with lower density (Damp104) is very slow (estimated wall time 6 months at frame 6.7 of 1500) and eventually crashed (timeout error)
- The more pronounced artefacts in Damp104 compared to Damp105 could be the reason, but it is not clear
- The larger temperature in Damp104 could mean that the computational unit TempScale~ K is too small. However, I've now played around a lot with the computational units and this does not make a significant difference.
- It is possible that AMR is causing the problem, so I'm now doing smaller box runs with and without AMR to see if it makes any difference.
Next Steps
- First try to fix this problem with low ambient densities
- We should use half the smoothing length (now done) but should we double the resolution? Should we increase the time for the relaxation run?
Meeting Update --08/03/17
- IOPP invoice for the OH231 paper
- XSEDE Resources
- Current resources: about 26000 Node-hour on stampede2 and 180000 cpu-hours on comet details
- Stampede2 currently only has KNL co-processors. details on this page
- Users
- has been working with Eric & Jason's students.
- Coding
- Updates to JetClump and pnStudy module: distribution of ambient density #438; total mass and momentum on the grids.
- OpenMP optimization for Common Envelop module: details on this page
- Wire Turbulence
- rearranged some figures and redid figure 1
- re-wrote introduction and method/model part. added more references
Update 8/3
- ATHENA and AstroBEAR comparisons are working nicely. There are still a lot of differences between the simulations, however. Dashed lines are ATHENA, solid are AstroBEAR.
Initial conditions
Flattening radius - ATHENA flattens at r_ib rather than r_mask
Electron densities - ATHENA goes to near zero inside planet, AstroBEAR is just 10-10 of the H density (this is probably not significant in the end)
r_ob - ATHENA profile extends to r_zero, but uses values at r_ob for density and pressure at edge
Pressure profiles differ
These last two seem likely to cause many of the differences seen once things actually start happening.
Frame 1
Looks like this ATHENA run was actually run with ramping - John seemed to believe it was not, so either the flag he's using isn't successful or he grabbed the wrong simulation accidentally
dE/dt for PdV, Lyman-alpha, and ionization heating all differ fairly significantly
AstroBEAR recombination is incredibly small - likely not attributable to differences in profiles
Other differences at this point are almost certainly attributable to the difference in flux or the difference in outer boundaries.
With the profile extended to rzero (and the resolution increased by one level), the location of the outer boundary becomes closer, but the values no longer match. So something is going wrong either in the calculation or import of the profile, still.
- Gave Diyora edge-centered VTK output from VisIt volume plot to start with. I almost have the layout working, but can't seem to get the refined data (only the coarsest is showing in the chombo). Is only level 0 data loaded BeforeGlobalStep?
- Did testing for OpenMP version of code. Problem appears sometime between 5 tasks/4 threads and 10 tasks/12 threads (haven't determined exact cutoff yet).
- Working on presentation for eclipse at Ogden Farmers' Library.
Recent progress
- Previously, we have this version of paper on the orbital period change in binary system with giant stars.
The referee asked us to show that our angular momentum evolution do not have serious problem. So I sampled the angular momentum through multiple of spherical shells in a single AGB star test in the co-rotating frame. Ideally, the angular momentum of the wind at any shell is the same, or not varying if we average of enough time. However, simulation has errors and sampling is never infinitely long.
We should also replace the five images of alpha in the appendix in the last version of the paper.
We are going to use two figures to show the corrected orbital period change.
exclude all angular momentum which measured at 0.8d in the AM error study
only exclude the spin angular momentum of the AGB star
I am rewriting the paper.
- I studied the Riemann problem with realistic EOS.
A study of EOS Riemann problem
The conclusion is, realistic gas behave somewhat like ideal gas with a smaller gamma, though not exactly the same. We can in the future run some simulation with gamma=1.1 instead of gamma=1.67.
I recommend to remodel the AGB wind and binary star model with this argument.