Вы находитесь на странице: 1из 34

Supersonic Turbulence, Intermittency, and AMR

Alexei Kritsuk
University of California, San Diego In collaboration with Rick Wagner, Mike Norman, & Paolo Padoan, UCSD

http://akpc.ucsd.edu/Isothermal

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Outline

 Motivation & Background  Intermittency and simulations with AMR  Turbulent structures at high Re  Statistical properties: AMR vs. uniform grid  Perspective  Summary

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Why adaptive mesh renement?


 Astrophysical motivation: Re 108 is needed to simulate turbulence in molecular clouds (MC) Hierarchical structure of MCs on scales from 10 pc to 0.01 pc. Dynamic range of > 106 (linear) is needed to correctly reproduce statistics of
self-gravitating prestellar cores.

AMR is indispensable to follow the evolution of gravitationally unstable prestellar cores,


their fragmentation and collapse down to formation of protostellar objects.
5 pc

0.02 pc

300 AU

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Pros & Cons

Frisch 1995: . . . the number of degrees of freedom could be signicantly smaller than Re9/4 , thereby brightening the prospects for numerical simulations (but not on
uniform grids).

NASA ATP Review Panel Consensus 2004: . . . it is not clear whether AMR is
the appropriate approach to simulate turbulence since high resolution is needed for structure throughout the entire computational domain.

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Should AMR work?


 Turbulence theory motivation: From K41 phenomenology, the resource required to simulate high-Re ows on a 3 uniform grid for a xed number of large eddy turnover times is: Nx Nt Re3 . Integral scale Reynolds number: Re Kolmogorov dissipation scale:
0 0 v0

1/4

3 v0 0

3 3 v3 0 0

1/4

Re3/4

However, turbulent ows are NOT completely chaotic on both large and small scales it is expected that NDoF < Re9/4 , but how much smaller? K41 does not include intermittency, but fully developed turbulence at high Re is intermittent. Dimension of the most singular dissipative structures D < 3: D = 1 for incompressible turbulence (vortex laments) D = 2 for supersonic compressible turbulence (shocks) Volume lling factor of boxes covering dissipative structures scales as 3D , where is the box size; N D (e.g. von Koch curve, D = 1.2619. . . ) 3D 1+D this makes it promising for AMR applications, since (ideally) NDoF Re
5 Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Our niche

 High-resolution simulations of supersonic turbulence


(Kritsuk, Norman & Padoan 2004, astro-ph/0411626, revised v2)

 Intermittency, dissipative structures, and scaling relations for


compressible turbulence

 Interplay of interstellar turbulence and star formation  Turbulence and hierarchical structure of molecular clouds in
the context of multiphase ISM
6 Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Our approach
 Input 3D Euler equations Periodic box; Cartesian mesh Quasi-isothermal EOS ( = 1.001) Driven turbulence (Mach 6, solenoidal driving force with a constant pattern) High-order Godunov-type method (PPM), strong shocks are two-zones-wide Uniform grids 643 , 1283 , 2563 , 5123 , 10243 with PPM Also 5123 and 10243 simulations with the ZEUS solver for comparison Structured AMR with 12 levels of renement by 4 where needed (shocks & shear)  Engine ENZO code for cosmology and astrophysics. Public since March 2004  Output Statistics of statistics
Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Sample tests with the PPM solver in ENZO


Noh Problem
http://cosmos.ucsd.edu/enzo

Implosion Test

Double Mach Reflection

Sedov Blast

KH Instability

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Time evolution of kinetic energy


Uniform grid simulations with resolutions 643 , . . . , 5123 were evolved for 6tdyn , t
PPM; M = 6; = 1.001; tdyn = 0.08 22 21 20 19 Ekin 18 17 16 15 14 0 0.1 0.2 0.3 time 0.4 0.5 0.6 1024AMR 1024 512 256 128 64

[0, 0.5].

1024AMR simulation with one level of renement by a factor of 4 was restarted from the 2563 data at t

= 0.4, and was evolved up to t = 0.5. = 0.4 from the


9

1024PPM and ZEUS (not shown) simulations were restarted at t

corresponding 5123 datasets by regridding to ner 10243 grids. Then both were evolved up to

t = 0.5.
Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Live structures

What you see depends on how you look at them...

Follow the web links below to watch MPEG animations of (projected) gas density:

Full box projection Thick slice (100 zones across) Thin slice (one zone across)

Smoke Mach cones Jellysh


[0.4, 0.5].

The animations are based on the 1024AMR run and cover the evolution for t

10

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

What is different at high resolution?


256PPM 1024ZEUS . 512PPM 1024PPM

11

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Structures: AMR vs. uniform grid


Density (log) slices from two simulations with the same effective resolution 256L1x4AMR 1024PPM

12

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Unrened regions within a slice


AMR subgrids cover approximately 65% of the computational volume.

13

Voids, as well as inactive windows within the high density regions, remain unrened. Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Substructure in large-scale dense laments

14

Large-scale shocks are fragmented, consistent with observations of molecular gas by Brunt (2003)
Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Nested U- and V-shapes in a thick density slice

15

Mach angles of large-scale structures in this image correspond to relative motion with M 2 3
Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Nested Mach cones in a thick density slice

16

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Statistics of statistics

PDFs, Power Spectra, and Structure Functions


All measurements are instantaneous at t

= 0.5, no time averaging is involved.

17

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Density PDF

0 -1 -2 log10 PDF -3 -4 -5 -6 -7 -3 -2 -1 0 log10 1

1024AMR 1024

18

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Density power spectrum

1024AMR 1024 slope = -0.90

-1

log10 P(k)

-2

19

-3

-4 0 0.5 1 1.5 log10 k 2 2.5 3

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Compensated velocity power spectrum

1024AMR 1024

1 Ev(k)

log10 k

5/3

20

0.5

1.5 log10 k

2.5

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Compensated velocity power spectrum

1024AMR 512

1 Ev(k)

21

log10 k 0 0

5/3

0.5

1.5 log10 k

2.5

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Compensated velocity power spectrum

1024AMR 256

1 Ev(k)

22

log10 k 0 0

5/3

0.5

1.5 log10 k

2.5

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Compensated velocity power spectrum

1024AMR 1024ZEUS

1 Ev(k)

23

log10 k 0 0

5/3

0.5

1.5 log10 k

2.5

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Compensated velocity power spectrum

1024AMR 1024 slope = -0.16

24
1 Ev(k) log10 k 0 0
5/3

0.5

1.5 log10 k

2.5

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

Transverse velocity structure functions (1024PPM) 10 p=1 2 3 4 5 6 7 8 9

25

log10 S (x)

0 -2 -1.5 -1 log10 x -0.5 0

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

Longitudinal velocity structure functions (1024PPM) 10 p=1 2 3 4 5 6 7 8 9

26

log10 S (x)

0 -2 -1.5 -1 log10 x -0.5 0

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

2nd order velocity structure functions (1024 PPM) 2 lngP 0.98 trvP 0.93

27

1.5

log10 S (x)

0.5

0 -2 -1.5 -1 log10 x -0.5 0

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

2nd order velocity structure functions (1024 PPM and ZEUS) 2 lngP trvP lngZ trvZ

28

1.5

log10 S (x)

0.5

0 -2 -1.5 -1 log10 x -0.5 0

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

3rd order velocity structure functions (1024PPM) 3 lng 1.31 trv 1.23

29

2.5

2 log10 S (x)
3

1.5

0.5

0 -2 -1.5 -1 log10 x -0.5 0

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

3nd order velocity structure functions (1024 PPM and ZEUS) 3 lngP trvP lngZ trvZ

30

2.5

2 log10 S (x)
3

1.5

0.5

0 -2 -1.5 -1 log10 x -0.5 0

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

31
Transverse velocity structure functions: 1024 PPM vs. AMR 8 7 6 5 4 3 2 1 0 -2 -1.5 -1 log10 x -0.5 0 PPM 1 2 3 4 5 6 7 AMR 1 2 3 4 5 6 7

Towards simulating turbulence with AMR

log10 S (x)

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Velocity structure functions

32
Longitudinal velocity structure functions: 1024 PPM vs. AMR 8 7 6 5 4 3 2 1 0 -2 -1.5 -1 log10 x -0.5 0 PPM 1 2 3 4 5 6 7 AMR 1 2 3 4 5 6 7

Towards simulating turbulence with AMR

log10 S (x)

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Perspective
Volume fraction covered by subgrids vs. their effective resolution.

33

An open diamond at the bottom is our forecast.


Towards simulating turbulence with AMR
CMSO Workshop Madison WI, June 20-21, 2005
c Alexei Kritsuk

Summary
 Numerical simulations of MC turbulence with AMR return reasonable results  For simulations of supersonic turbulence [with a high-order Godunov scheme], AMR is benecial at resolutions 109  High Re[solution] is needed to reproduce a complex pattern of nested hierarchical dense
structures in supersonic turbulent ows

34

 These structures originate in (non)linear instabilities: Kelvin-Helmholtz instability [Helmholtz 1868] Nonlinear supersonic vortex sheet instability [Miles 1957] Nonlinear thin shell instability [Vishniac 1994] ...  The instabilities, thus, may control the scaling properties of turbulence  The fractal dimension of the dynamically important structures, 2 < D 2.3, is consistent
with observations [e.g., Elmegreen & Falgarone 1996; Chappell & Scalo 2001]

Towards simulating turbulence with AMR

CMSO Workshop Madison WI, June 20-21, 2005


c Alexei Kritsuk

Вам также может понравиться