# Cosmological Simulations with Adaptive Smoothed Particles Hydrodynamics

**ABSTRACT** We summarize the ideas that led to the Adaptive Smoothed Particle Hydrodynamics (ASPH) algorithm, with anisotropic smoothing and shock-tracking. We then identify a serious new problem for SPH simulations with shocks and radiative cooling --- false cooling --- and discuss a possible solution based on the shock-tracking ability of ASPH.

**0**

**0**

**·**

**0**Bookmarks

**·**

**34**Views

- Citations (0)
- Cited In (0)

Page 1

arXiv:astro-ph/0110038v1 1 Oct 2001

Astrophysical Supercomputing Using Particles

IAU Symposium, Vol. 208, 2001

J. Makino and P. Hut, eds.

Cosmological Simulations with Adaptive Smoothed

Particle Hydrodynamics

Hugo Martel

Department of Astronomy, University of Texas, Austin, TX 78712, USA

Paul R. Shapiro

Department of Astronomy, University of Texas, Austin, TX 78712, USA

Abstract.

Particle Hydrodynamics (ASPH) algorithm, with anisotropic smoothing

and shock-tracking. We then identify a serious new problem for SPH

simulations with shocks and radiative cooling — false cooling — and

discuss a possible solution based on the shock-tracking ability of ASPH.

We summarize the ideas that led to the Adaptive Smoothed

1. Introduction

SPH is the most widely used numerical method for cosmological simulations

with gas dynamics. With our collaborators, we have developed a new version of

SPH, called Adaptive SPH (ASPH), which addresses some specific limitations of

standard SPH (Shapiro et al. 1996; Owen et al. 1998; Martel & Shapiro 2002).

For a given number of particles, ASPH resolves much better than standard SPH

whenever anisotropic collapse or expansion occurs. ASPH also has a shock-

tracking algorithm that can be used to restrict spurious heating by artificial

viscosity. ASPH simulations in 3D of explosions during galaxy formation are

described elsewhere (Martel & Shapiro 2000, 2001a, b).

2. Standard SPH vs. Adaptive SPH (ASPH)

For kernel smoothing in standard SPH to be accurate, the smoothing length

h must be a few times the mean particle spacing ∆x.

variations in the fluid quantities are oversmoothed, leading to a loss of resolution;

if h < ∆x, particles lose contact with their nearest neighbors, resulting in a loss

of accuracy. Since the mean particle spacing varies with time and space, each

particle must carry its own smoothing length, which varies with time to reflect

the expansion or contraction of the fluid. This is illustrated in Figure 1. The

top row shows a 2D distribution of particles contracting isotropically. We focus

on one particle, shown by a large dot. The smoothing length h of that particle

defines a zone of influence, indicated by a circle. As the fluid contracts, h is

reduced in proportion to the mean particle spacing. The second row illustrates

the case of an anisotropic contraction, the planar collapse of a sinusoidal density

perturbation, leading to the formation of a caustic. In the direction of collapse,

If h ≫ ∆x, spatial

1

Page 2

2

Hugo Martel & Paul R. Shapiro

Figure 1.

evolution of the H-tensor.

by a large dot.

Top row: Isotropic contraction with isotropic smoothing.

row: Anisotropic contraction with isotropic smoothing. Bottom row:

Anisotropic contraction with anisotropic smoothing.

Contraction of 2D particle distributions, illustrating the

We focus on one particle, represented

The solid curve represents the zone of influence.

Middle

Page 3

Adaptive Smoothed Particle Hydrodynamics

3

the smoothing length does not shrink fast enough to keep up with the contraction

of the fluid, and eventually greatly exceeds the mean particle spacing, leading to

poor resolution. In the transverse direction, the smoothing length shrinks but

the fluid does not contract, and horizontal rows of particles progressively lose

contact with one another. The third row illustrates the ASPH approach. The

smoothing is anisotropic. The smoothing length is replaced by an H-tensor which

defines an elliptical zone of influence (ellipsoidal in 3D). This zone of influence

deforms and rotates to follow the deformation of the fluid.

length is now direction dependent, and remains proportional to the mean particle

spacing in any direction.

Artificial viscosity is necessary to allow the formation of shocks, but can

lead to spurious preheating of gas contracting supersonically far from any shock.

In ASPH, the evolution of the H-tensors can be used to track the location of

shocks and apply artificial viscosity selectively. This is illustrated in the third

row of Figure 1. In ASPH, viscous heating is initially turned off. As the system

evolves, the smoothing ellipsoid flattens, with one axis approaching zero length.

ASPH uses this to determine when a particle is about to be shocked and turns

viscous heating on for that particle. This restricts viscous heating to particles

encountering shocks, as needed.

To illustrate how ASPH achieves higher resolution than standard SPH, we

focus on a stringent test, the planar collapse of a sinusoidal plane-wave density

perturbation, the cosmological pancake problem. The fluctuation grows from

linear to nonlinear amplitude and forms a caustic in the dark matter distribution

in the plane of symmetry at scale factor a = ac, with strong accretion shocks

located on each side of the central plane. We use equal-mass dark matter and

gas particles, in a universe with Ω = 1 and Ωgas= 0.5, with 64 particles per

pancake wavelength λp, to evolve the system to a = 2.333ac. Figure 2 shows gas

density, temperature, velocity, and pressure profiles (in computational units) at

the final time. The shock is located at log(x/λp) = −1.3 (from midplane). The

ASPH results are significantly better than the SPH results, which is especially

noticeable in the postshock density and temperature profiles. The ASPH profiles

follow the exact postshock solution over nearly 3 orders of magnitude in length,

while the SPH profiles level off because of limited spatial resolution.

The smoothing

3.False Cooling

A serious numerical problem emerges for both algorithms when radiative cooling

is added. In Figure 3, we plot a typical cooling function commonly used in

cosmological simulations. Let us assume that the rectangle in the top panel of

Figure 3 represents the physically relevant region. The cooling rate drops sharply

at low and high temperatures. To approximate this behavior, we consider a

simplified cooling window function illustrated in the bottom panel of Figure 3.

The cooling rate is constant inside a fixed temperature window (Region II), and

zero outside that window (Regions I and III).

In principle, when a fluid element is shock-heated from Region I to Re-

gion III, it should cross the shock too fast to cool radiatively as it passes thru

Region II. The numerical shock has a finite thickness which is unphysically large,

however, so the shock transit time, tshock, can artificially exceed the physical

Page 4

4

Hugo Martel & Paul R. Shapiro

Figure 2.

sity profile. Top right panel: temperature profile. Bottom left panel:

velocity profile. Bottom right panel: pressure profile. Solid curves,

crosses, and circles show the exact, standard SPH, and ASPH results,

respectively, with distances in units of λp, the pancake wavelength.

The inset in the top right panel shows an enlargement of the immedi-

ate postshock region.

Pancake collapse at a/ac= 2.333. Top left panel: gas den-

Page 5

Adaptive Smoothed Particle Hydrodynamics

5

Figure 3.

in cosmological simulations. Bottom panel: top-hat cooling function

used to approximate the region of the cooling curve indicated by the

rectangular box in the top panel.

Top panel: typical radiative cooling rate commonly used

cooling time in Region II, tcool, causing spurious radiative cooling. Since tshock

scales like the shock thickness divided by the preshock velocity, it depends on

resolution, as the number of particles across the shock tends to be constant. If

the cooling rate is so large or the resolution so poor that tshock> tcool, particles

will be unable to cross Region II, as cooling forces them back to Region I, and

the shock does not form. We call this the False Cooling Problem.

To illustrate this problem, we introduce a window cooling function to the

cosmological pancake problem. Figure 4 shows the temperature and pressure

profiles at a = 2.333ac.The shaded area indicates the cooling window. If

cooling is slow (left panels), the shocked gas particles are able to cross the cooling

window and accumulate correctly in the postshock region. The particles located

at the bottom of the cooling window are those whose postshock temperature

was inside the cooling window. They cooled after having been shocked. When

cooling is fast (right panels), however, particles are incapable of reaching their

correct postshock temperature, and the shock does not even form.

4.Possible Solution

The false cooling problem only affects particles that are going through a shock

transition. If the algorithm could track the location of shocks, then false cooling

could be eliminated, simply by not allowing particles undergoing shock transi-

Page 6

6

Hugo Martel & Paul R. Shapiro

Figure 4.

cake collapse with radiative cooling. Shaded area indicates location of

cooling window. Left panels: slow cooling. Right panels: fast cooling.

Solid curves, crosses, and circles show the correct solution [by high-

res., Lagrangian hydro method of Shapiro & Struck-Marcell (1985)],

the standard SPH results, and the ASPH results, respectively.

Temperature and pressure profiles at a/ac= 2.333 for pan-

Page 7

Adaptive Smoothed Particle Hydrodynamics

7

Figure 5.

pancake collapse with fast cooling, using ASPH. Shaded area indicates

the location of the cooling window. Left panels: no cooling while a

particle shock-heats. Right panels: no cooling while a particle shock-

heats, and pseudo-conduction added.

Temperature and pressure profiles at a/ac = 2.333 for

Page 8

8

Hugo Martel & Paul R. Shapiro

tions to cool. There is already a shock-tracking algorithm in ASPH, which is

used to restrict viscous heating. The algorithm identifies particles located in

shock transitions and turns viscous heating on for them. We can attempt to

solve the false cooling problem by turning cooling off for the same particles,

with results as shown in the left panels of Figure 5. Particles are now able to

reach the postshock temperature and pressure (compare with the right panels

of Fig. 4), and the ASPH solution is close to the exact one. There are some

oscillations near the contact discontinuity between hot and cooled gas, but since

it involves only a few particles, these oscillations are underresolved and do not

feed back into the evolution of the system. The right panel shows an attempt

to improve the solution by introducing a small amount of artificial conduction.

The oscillations are gone, but the postshock profile is not as well reproduced.

5. Summary

Anisotropic smoothing kernels enable the ASPH algorithm to simulate prob-

lems involving strongly anisotropic collapse or expansion more accurately than

isotropic SPH, with better length resolution for the same number of particles.

Additional improvement results because ASPH tracks the location of shocks and

restricts viscous heating to particles overtaken by shocks.

We have identified a major problem, called false cooling, which can affect

cosmological simulations whenever the cooling rate has temperature peaks. False

cooling prevents shocked gas from reaching regions of high temperature where

the cooling is low. Solving this problem may require a shock-tracking capability

like that of ASPH. Our preliminary results are encouraging.

Acknowledgments.

dro simulations. This work was supported by TARP Grant 3658-0624-1999 and

NASA Grants NAG5-10825 and NAG5-10826.

We thank Ilian Iliev for help with our Lagrangian hy-

References

Martel, H., & Shapiro, P. R. 2000, Nucl.Phys.B, 80, 09/11 (astro-ph/9904121)

Martel, H., & Shapiro, P. R. 2001a, Rev.Mex.A.A. (SC), 10, 101 (astro-ph/

0006309)

Martel, H., & Shapiro, P. R. 2001b, in Proceedings of the 20thSymposium on

Theoretical Astrophysics, eds. J. C. Wheeler & H. Martel (AIP), in press

(astro-ph/0104068)

Martel, H., & Shapiro, P. R. 2002, in preparation.

Owen, J. M., Villumsen, J. V., Shapiro, P. R., & Martel, H. 1998, ApJS, 116,

155

Shapiro, P. R., Martel, H., Villumsen, J. V., & Owen, J. M. 1996, ApJS, 103,

269

Shapiro, P. R., & Struck-Marcell, C. 1985, ApJS, 57, 205

#### View other sources

#### Hide other sources

- Available from Hugo Martel · Jan 29, 2013
- Available from arxiv.org