Embedded disks in the AGB wind. I've ran two sims:

1. a disk contained in the orbital plane,
   • log density movie (upper part of the orbital plane, initially the disk is on the left and the AGB on the right): ​http://www.pas.rochester.edu/~martinhe/2011/binary/binary-n-disk1-dens.gif
   • vx movie (same display setup as the dens movie above): ​http://www.pas.rochester.edu/~martinhe/2011/binary/binary-n-disk1-vx.gif

I've only glanced at these movies and data, so the following are simple observations which need some serious thought.

• The structure of the disk is quickly lost
• The disk's former gas evolves and adoptes a structure which looks like the tilted disks of the binary sims (see table below).
• There is a shock wave coming from the AGB. This happened because I didn't adjust the density to match the one corresponding to the low resolution I used for this test. Yet, seems the diks looses its structure before the shock reaches the secondary's radius. I'll fix this for future tests, if any.

2. a disk with an ang. mom. vector at an angle of 30^o with the orbital ang. mom. vector.
   • log density movie (upper part of the orbital plane, initially the disk is on the left and the AGB on the right): ​http://www.pas.rochester.edu/~martinhe/2011/binary/binary-n-disk2-dens.gif
   • vx movie (same display setup as the dens movie above): ​http://www.pas.rochester.edu/~martinhe/2011/binary/binary-n-disk2-vx.gif

29 Mar '12

a=40 AU tests.

---

About run speed. It is proportional to:

1. the resolution

1.1 dx_max / dx_min

2. the filling ratios. They are not crucial for these sims because I'm using particle refinements, not to the AMR (which will not really help to seed up the runs). Yet, I control the radii of cells of each level, taken form the particles' center. I've adjusted these to be small enough to aid the run speed but large enough to capture the AGB wind and the Bondi radius.
3. dt, which is proportional to cfl*dx_min/ max(v_w, c_s, v_orbit)
4. the No. of processors used, N_p
5. cluster communication and processors' clock speed
6. the advance speed (computation time per grid level).

My tests indicate that for separations ≥ 20AU, the resolution should be >~ .047AU (64³+5refs) to form disks. I consistently see that bhive's runs are faster than bgene's (even when I use four times more processor in the latter than in the former). What worries me is the advance speed. The code's output shows that the simulation spends a good deal of time in this process. e.g. test 6 (column 6 in the table below) shows advances between the levels 4, 5 and 6 (which go back and forth several times between each level 0 dt) take as long as 12 secs. So every full timestep advance takes ~minuts to happen. Why, and how can I improve this?

---

r_B=2Gm₂/(v_w² + c_s² + v₂²) [from ^*],

r_B^'=Gm₂/(v_w² + c_s²) ,

where r_B, m₂, v_w, c_s, and v₂ are the Bondi radius, the secondary's mass, the sound speed and the secondary's orbital velocity with respect to the center of mass.

Test	q=m1/m2	resolution	rsoft/dx	rB/dx	rB/rsoft	rB'/dx	rB'/rsoft	Tw [K]	vw [km/s]	M'w [10-5Mo/yr]	time [orb]	log(dens/cu),vel/Mach
1 (bgene, "Rsoft8")	1.2/.6=2*	642x16+6refLev, dx=.023AU	2	448.7	224.3	272.1	136	300*	15	1	.016
2 (bgene, "2", running)	1.2/.6=2*	1282x32+6refLev, dx=.011AU	2	448.7	224.3	544	272.1	300*	15	1	.002
3 (bhive)	1.2/.6=2*	642x16+7refLev (slightly larger grid), dx=.015AU	6	718	119.6	435.3	72.6	300*	15	1	.102
4 (bhive, "2", running)	1.2/.6=2*	1282x32+5refLev, dx=.023AU	3	448.7	149.6	272.1	90.7	300*	15	1	.061
5 (bgene, "3", running)	1.5/1=1.5+	1282x32+5refLev, dx=.023AU	2	693	346.6	424.1	212.1	1000	15	10	.008
6 (bgene, "4", running)	1.5/1=1.5+	1282x32+6refLev, dx=.011AU	2	1386.4	693.1	848.2	424.1	1000	15	10	.001	< tilt
AGAIN 6 (bgene, "4", running)											.0027a
7 (bhive, "3", running)	1.5/1=1.5+	642x16+5refLev, dx=.047AU	4	301.1	75.3	179	44.7	3000	9	2	.062

^* val-Borro et al.

⁺ M&M '98

^a ​http://www.pas.rochester.edu/~martinhe/2011/binary/gene-4.gif

---

20 mar '12

As in M&M '98, model 1. I did a similar test before (see 21 feb '12 post), but it was shorter, with less resolution and with a separation

• Temp=3000K
• AGB mass-loss = 10^-5 M_o yr^-1, from time=0
• vel_wind=9 km/s, from time=0
• a=40AU
• circular orbit
• q=1.5 (m₁=1.5; m₂=1 M_sun)
• r_soft=2
• 64x64x16cells + 5 particle grid refinements
• grid: x,y ⇐|32AU|; z⇐|8AU|.

Density iso-contours, 360o look at the disk disk, time=.5 orbits, zoom in. Tilted structure formed by wind capture.

​http://www.pas.rochester.edu/~martinhe/2011/binary/20mar1144.gif

Running the same model but with r_soft=0.

---

As in Val-Borro et al. '09 (​http://adsabs.harvard.edu/abs/2009ApJ...700.1148D)

• Temp=1000K
• AGB mass-loss = 10^-5 M_o yr^-1, from time=0
• vel_wind=10 km/s, from time=0
• a=40AU
• circular orbit
• q=2 (m₁=1.2; m₂=.6 M_sun)
• r_soft=2
• 64x64x16cells + 5 particle grid refinements
• grid: x,y ⇐|60AU|; z⇐|15AU|.
• Running time (24 afrank p, bluehive) = 4 days

Log(dens) on orbital plane

Density iso-contours, 360o look at the disk disk, time=1 orbits, zoom in. Strongly tilted disk.

​http://www.pas.rochester.edu/~martinhe/2011/binary/40au-bb5-3d.gif

Here are some snapshots of the orbital plane velocity field, ​http://www.pas.rochester.edu/~martinhe/2011/binary/40au-bb5-vel.pdf

Here are some plots of the mean angular momentum in the grid as a function of time, ​http://www.pas.rochester.edu/~martinhe/2011/binary/mAngMom.png

28 feb '12

As in M&M '98 model 3:

• Temp=2400K
• AGB mass-loss = 10^-5 M_o yr^-1 (they actually use 8x10^-6)
• a=17AU
• vel_wind=15 km/s
• circular orbit
• q=1.5
• r_soft=2
• 128x128x64cells + 4 particle grid refinements
• grid: x,y ⇐|25AU|; z⇐|12.5AU|; runs faster. I've checked and there's no boundary inflow.

Density iso-contours, edge-on, time=.5 orbits, zoom in. No tilting. The captured wind does show a disk-like shape, but at this point the disk is forming.

24 feb '12

Same model as yesterday, time=2 orbits. There's some inflow from all the boundaries onto the grid (left panel, small v_z from all boundaries), so I need a larger, or a cubic, grid and/or wind objects to block the inflows. Easy to solve. The inflow makes the flow patten quite messy and destroys the spiral structure induced by the orbit (middle panel, orbital plane view). The disk is not strongly affected by the inflow, which is good (right panel, low-res perspective view of iso-contours showing that the disk is plane symmetric and has an asymmetric bow shock which faces the AGB star).

I'll stop this simulation and send another one with the same setup but: a cubic grid and outflow-only wind objects at the boundaries.

23 feb '12

A very handsome disk!!'''

• Temp=3000K (as in M&M '98)
• AGB mass-loss = 9.9x10^-6 (as in M&M '98)
• a=3.7AU < M&M '98
• vel_wind=9 km/s (as in M&M '98)
• circular orbit
• q=1.5 (as in M&M '98)
• r_soft=2
• 128x128x64cells + 4 particle grid refinements⁺
• grid: x,y ⇐|1|; z⇐|.5|. Runs faster. We can use the full cubic grid for production runs.

Next steps:

1. try M&M '98 model 4.
2. elliptical orbit

⁺ Runs faster (1 orbit per 16 running hrs) than 32x32x16+6 particle refs (based on another test using 32³ cells +5 particle refinements which produced 1 orbit per 17 running hrs; yet for the latter test I used r_soft=0 which produces high velocities in the central-most orbits of the disk, hence reducing the timstep); the more grid refinements I request, the grater the numer of sub-steps the code has to perform. This is for 48 afrank processors.

22feb '12

Time=2.7orbits, no tilt.. The secondary keeps accreting mass and capturing wind material. I still see significant density gradients in the bound gas. Mastrodemos & M do not see this, so I think this is purely a shock effect. De Val-Borro et al. do see something similar, specially for their large wind region models (​http://adsabs.harvard.edu/abs/2009ApJ...700.1148D), but this may be a 3D effect. This image shows 5 flow steam lines of the wind captured gas.

I want to run this problem again with 1 more refinement level and a r_soft=2dx which should reduce the velocity of the central most cells. I presume this will make the disk radius larger.

21 feb '12

As in M&M '98 model 1:

• Temp=3000K
• AGB mass-loss = 9.9x10^-6
• a=3.7AU
• vel_wind=9 km/s
• circular orbit
• q=1.5
• 32³ + 5 particle grid refinements
• r_soft=0

I see a disk-looking structure (with a larger radius than the one in the previous test), with a Keplerian-like vel distribution and a flow structure that's symmetric with respect to the orbital plane (i.e. no tilting), after~1.3 orbits. The simulation is still running. I want to see it at 5 orbits.

Log density in grayscale. Velocity field in color scale in Mach units.
Zoom in to the disk

Central disk + Bondi:

I do not see tilting in the central region of the disk. Note that for this setup (taken from test 2*) rsoft=4dx, hence the particularly soft-looking center of the disk. This is about 25% of the run time that I used in test 2*, but it is well within the tilting growing time.

^* ​https://clover.pas.rochester.edu/trac/astrobear/blog/martinhe02032012

---

20 feb '12

As in M&M '98:

• Temp=2400K
• AGB mass-loss = 9.5x10^-6
• a=12AU (they actually have 12.6 AU)
• vel_wind=25km/s
• circular orbit
• q=1.5
• 32³ + 4 particle grid refinements
• r_soft=0

I see a disk-looking structure, although smaller than I expected, after~7.5 orbits. Again, it is symmetric with respect to the orbital plane, hence Jonathan's Bondi accretion implementation see to have solved the tilting enigma. I need more resolution to capture the full formation of the disk.

15 feb '12

Same as below but 64³ + 5 particle grid refinements and temp_wind=1000K (and in an exploratory elliptical orbit)

PRELIMINARY: I see a disk-looking structure after~3.7 orbits which is symmetric with respect to the orbital plane!!

14 feb '12 Jonathan has implemented, and tested with a central disk problem setup, the Bondi accretion of sink particles. I've been trying it in one instance of my binary-formed disks problem with:

• isothermal solver, gamma= 1.001
• r_soft=0 (as suggested by Jonathan)
• a=20 AU
• q=1.5
• vel_wind=20km/s; temp_wind=2000K (line M&M '97); mass-loss_wind=10^-5 M_sun /yr
• 64³ + 3 particle grid refinements, so dx~2.1 AU
• r_Bondi=4 AU; r_Bondi/r_soft=NA

I do not see a disk even after 20 orbits. The wind captured structure does look different than the one formed without* the Bondi accretion. I.e. it shows a light core and a dense tail (see left and upper right panels), but not the other way around (as in * ). Also, the structure is symmetric with respect to the orbital plane (see lower right panel).

I'm now trying this setup, but with these changes:

• 5 particle grid refinements ( not 6 yet for it will be rather slow), so that r_Bondi=8dx (instead of 2dx)
• temp_wind=1000K, instead of 2000.

^* see test 1 in ​https://clover.pas.rochester.edu/trac/astrobear/blog/martinhe01292012