+ All Categories
Home > Documents > How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE,...

How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE,...

Date post: 17-Jan-2016
Category:
Upload: clara-gaines
View: 215 times
Download: 0 times
Share this document with a friend
Popular Tags:
28
How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger (La Sapienza, Rome) A. Gabrielli (Istituto dei Sistemi Complessi-CNR , Rome) B. Marcos & F. Sylos Labini (Centro E. Fermi, Rome)
Transcript
Page 1: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

How accurately do N body simulations reproduce the clustering of CDM?

Michael Joyce

LPNHE, Université Paris VI

Work in collaboration with:

T. Baertschiger (La Sapienza, Rome)A. Gabrielli (Istituto dei Sistemi Complessi-CNR , Rome)

B. Marcos & F. Sylos Labini (Centro E. Fermi, Rome)

Page 2: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Outline

Intro:Theory vs. simulation or what is the problem?Qualitative expectations or is it a real problem?

Systematic analytical approaches: (Initial conditions)Perturbative regime

Towards control of the non-linear regimeComments on numerical testingOther approaches

Page 3: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

What is the problem?

N body simulations

are not a direct discretization

of the theoretical equations of motion

a “numerically perfect” simulation ≠ theory

Page 4: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

What is the problem?

Theory (What one would like to simulate)

Purely self-gravitating microscopic particles (typically ~1070/[Mpc]3)

Treated statistically ---> Vlasov-Poisson equations for f(v,x,t)

“Collisionless” (mean field) limit

Fluid/continuum limit (appropriate N --> )

Physics: separation of scales

scales of discrete “graininess” << scales of clustering

Page 5: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

What is the problem?

N body systems (What is in fact simulated)

Purely self-gravitating macroscopic particles (typically ~(1-100)/[Mpc]3)

Direct evolution under Newtonian self-gravity

Expanding background + small scale smoothing on force

Page 6: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

What is the problem? The discreteness (finite N) problem

What is the relation between e.g. a correlation function or power spectrum

calculated from output of an NBS and the same quantity in the theory?

Answer: we don’t know !

Since theory is an appropriate N --> limit, the problem may be stated:

what are the corrections due to the use of finite N ?

Page 7: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Is it really a problem?

Is N ~ 1010 (e.g. “Millenium”) not enough?

Answer: it depends on what you want to resolve.

Simulators systematically make very optimistic assumptions

Surely simulators understand and control this?

Answer: No!

There are some (but very few) numerical studies.

In general only qualitative arguments for trusting results are given.

Page 8: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Is it really a problem? The issue of resolution

Unphysical characteristic scales are introduced by the “discretization”:

Interparticle separation l, force smoothing [and box size L, with N=(L/l)3]

Naively: fluid continuum limit for scales >> l

In practice: results are taken as physical (usually) down to , where << l

Why? This is the “interesting” regime (strongly non-linear)…

e.g. “Millenium” simulation: l ≈ 0,25 h-1 Mpc, ≈ 5 h-1 kpc

Is it justified? If so, what are errors?

Page 9: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Is it really a problem? Some common wisdom justifying this practice

Numerical tests show that results are robust to changes in N (---> l)

Some analytical “predictions” work well: notably

Press-Schecter formalism

Self-similar scaling for power law initial power spectra

Physics: “transfer of power to small scales is very efficient”

Page 10: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Is it really a problem? Caveats to this common wisdom

Numerical studies in the literature are

few and unsystematic (other parameters varied --- see below),

very limited range of l (at very most factor of 10, typically by 2)

do not agree (e.g. Melott et al.conclude that extrapolation is not justified)

Physics: PS, self-similarity --> structures form predominantly by collapse, with

linear theory setting the appropriate mass/time scales.

This does not establish validity of Vlasov/fluid description in non-linear regime.

Important: N independence does not imply Vlasov!

Page 11: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Is it really a problem? So..

Our understanding of this fundamental issue about NBS is, at best, qualitative

We need a “theory of discreteness errors” leading to:

A physical understanding of these effects

Methods for quantifying these effects (analytically or numerically)

Page 12: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Rest of talk: A problem in three parts

Initial conditions of simulations

The perturbative regime (up to “shell-crossing”)

The non-linear regime

Page 13: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Analytical approaches I Discreteness effects in initial conditions (IC)

IC are generated by displacing particles off a lattice (or “glass”) using

Zeldovich Approximation.

Page 14: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.Input theoretical power spectrum

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.

QuickTime™ and aTIFF (LZW) decompressorare needed to see this picture.Convolution term (linear in Pth)

power spectrum of lattice (or glass)

Analytical approaches I Full power spectrum of discrete IC

Page 15: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.
Page 16: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Theoretical correlation properties very well represented in reciprocal space for k<kN with discrete contribution at k>kN

In real space (e.g. mass variance) the relation is more complicated (discreteness terms are delocalized)--->In the limit of low amplitude (i.e. high initial red-shift), atfixed N, the real space properties are not represented accurately

Is this of dynamical importance?

Analytical approaches I Conclusions on discreteness in IC

Page 17: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Evolution of N body system can be solved perturbatively in displacements off the lattice

Gives discrete generalisation of Lagrangian perturbative theory for fluid.

---> Recover the fluid limit and study N dependent corrections to it

Analytical approaches II

Perturbative treatment of the N body problem

Page 18: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Analytical approaches II

Linearisation of the N body problem

Page 19: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Analytical approaches II

Linear evolution of displacement fields

Page 20: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Analytical approaches II

Eigenvalues for a simple cubic lattice

Page 21: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Analytical approaches II

Growth of power in “particle linear theory”

Page 22: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Title: aniso.agr

Creator: Grace-5.1.20

Preview: This EPS picture was not saved with a preview (TIFF or PICT) included in it

Comment: This EPS picture will print to a postscript printer but not to other types of printers

Analytical approaches II

Corrections in amplification due to discreteness

• Simulation begins at a=1 • Deviation from unity is the discreteness effect

Page 23: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Analytical approaches II

What we learn from this perturbative regime

Fluid evolution for a mode k recovered for kl << 1 i.e. as naively expected.

Exact fluid evolution is thus recovered by imposing a cut-off kC in the input

power spectrum, and taking kC l --> 0

Discreteness effects in this regime accumulate in time.

Taking initial red-shift zI -->, at fixed l, the simulation diverges from fluid

(--> zI is a relevant parameter for discreteness!)

These dynamical effects of discreteness are not two-body collision effects

Page 24: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Not analytically tractable (that’s why we use simulations!)

Need at least well defined numerical procedures to quantify discreteness

Some approaches towards understanding physics:

Detailed study of “simplified” simulations (e.g. “shuffled lattice”)

Rigorous studies of simplified toy models

(--> statistical physics of long range interactions)

Towards control on the non-linear regime

Page 25: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Increasing N to test for discreteness effects we should extrapolate towards the correct continuum limit.

Formally it is N --> i.e. l --> 0 (in units of box size)

What do we do with other relevant parameters: , zI, kC ?

(Non-unique) answer: keep them fixed (in units of box size for , kC )

Note: For robust conclusions on NBS we need to extrapolate to l << kC-1

l<< ---> large PM type simulations

Towards control on the non-linear regime

The continuum limit

Page 26: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

Lattice with uncorrelated perturbations (<--> random error on

positions)

Power spectrum k2 at small k

Non-expanding space

Findings:

Self-similarity with temporal behaviour of fluid limit

Form of non-linear correlation function already defined in nearest

neighbour dominated (i.e. non-Vlasov) phase.

N body “coarse-grainings” only converge in continuum limit (as above)

Towards control on the non-linear regime

Study of “shuffled lattices”

Page 27: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

N body simulators make very optimistic and rigorously unjustified

assumptions about extrapolation to theory

New formalism resolving the problem in the perturbative regime

(--> defined continuum limit, quantifiable error, “correction” of IC)

Physical effects of discreteness are more complex than two body

collisionality + sampling in IC

Numerical tests should extrapolate to continuum limit as defined.

Other numerical and analytical approaches necessary.

Conclusions

Page 28: How accurately do N body simulations reproduce the clustering of CDM? Michael Joyce LPNHE, Université Paris VI Work in collaboration with: T. Baertschiger.

References

M. Joyce, B. Marcos, A. Gabrielli, T. Baertschiger, F. Sylos Labini Gravitational evolution of a perturbed lattice and its fluid limitPhys. Rev. Lett. 95:011334(2005)

B. Marcos, T. Baertschiger, M. Joyce, A. Gabrielli, F. Sylos Labini Linear perturbative theory of the discrete cosmological N body problemPhys.Rev. D73:103507(2006)

M. Joyce and B. Marcos, Quantification of discreteness effects in cosmological N body simulations. I: Initial conditionsPhys. Rev. D, in press,(2007)

M. Joyce and B. Marcos, Quantification of discreteness effects in cosmological N body simulations. II: Early time evolution.In preparation (astro-ph soon)

T. Baertschiger M. Joyce, A. Gabrielli, F. Sylos Labini Gravitational Dynamics of an Infinite Shuffled Lattice of Particles Phys.Rev. E , in press (2007)

T. Baertschiger M. Joyce, A. Gabrielli, F. Sylos Labini Gravitational Dynamics of an Infinite Shuffled Lattice: Particle Coarse-grainings, Non-linear Clustering and the Continuum Limit , cond-mat/0612594


Recommended