A bit of (my) history My main PhD

simulations were performed on COSMOS Mk I in 1998-99!

32 R10000, 8 GB of memory, $2,000,000 0.5×106 particles, only

4,000 timesteps Simulations I’ll talk

about today, 32 core servers, with 64 GB, $20,000 2.5×106 particles, but

×10 timesteps

AGN feedback modelling: a

comparison of methods

(a work in progress) Rob ThackerAssociate Professor

& Canada Research Chair

Saint Mary’s University, Canada

Credit where a lot of credit is due

This work is part of PhD student James Wurster’s thesis

Outline Motivation

Physics issues, obs vs theory Methods

Difficult choices to make, complicating factors

Problem(s) and resolution(s) Our results Conclusions

In a PhD thesis, far, far away….

Motivation Obs. evidence of AGN feedback has

been noted for years Is the observational case compelling?

Schawinski et al 2007, Fabian review (arXiv:1204.4114)

Large ellipticals case is pretty good Radio mode commonly observed

Still need to understand situation in intermediate masses, plus redshifts

Feedback Terminology Radio mode Accreting hot

gas Sub-Eddington

luminosity Radiatively

inefficient accretion

Radio jets provide heat source

Quasar mode Accreting cold

gas Up to Eddington

luminosity Radiatively

efficient accretion disk

Why compare? Comparison studies:

1999 Santa Barbara cluster comparison 2006 Radiative transfer comparison 2011 Aquila galaxy formation

comparison Don’t give any real “answers”

But do provide estimates of variation between methods

=> “Be careful” about results until 3 groups agree on it

Remember…

“The 9 orders of magnitude in physical scale means that all such simulations include subgrid assumptions and approximations.”

- Andy Fabian

The Optimistic Numericists view:

Can we be “unwrong” enough to give good insight?

Some thoughts to ponder…

Timescale between onset of nuclear inflow and AGN activity ~ 108 yrs

Many dynamical signatures evolve signifcantly on that time scale

ALMA + JWST will be an enormous help Simultaneous SFRs, mass inflow rates,

understanding radiative behaviour Good reasons to be optimistic

Prototype merger

Merger movie

Four base models + one extra

Springel, di Matteo, Hernquist 2005

(SDH05)

Okamato, Nemmen & Bower 2008

(ONB08)

Booth & Schaye 2009 (BS09, slightly

odd one out)

De Buhr, Quataret, & Ma 2011(DQM11)

+WT2012

But plenty of other work is related:

High res simulations of individual BH evolution/small scale

accretion

e.g. Levine et al 2008, 2010Alvarez, Wise & Abel 2009

Kim et al 2011Hopkins & Quateart 2010

Other “collision” work

e.g. Johansson, Naab & Burkert 2009

Halo evolutione.g. Sijacki et al 2009

Five key components Model for BH accretion rate

(Feedback) energy return

algorithm

SPH particle accretion algorithm

Black hole advection algorithm

Black hole merger

algorithm

Accretion physics Accretion of gas on to point in 1d:

Bondi-Hoyle-Lyttleton (1939,1944,1952) 

 

 

- Gas density & sound speed at infinity

- Velocity of BH wrt to (distant) gas

Accretion physics II Maximal symmetric accretion rate is

limited by the Eddington rate 

- Proton mass and Thompson X-section

- Efficiency of mass to energy conversion

 

 

Problems with BHL Physics:

2d problem is known to produce unstable flow

Material inflow not radial – what about angular momentum?

Radiative, magnetic effects etc Numerics:

How to relate physical variables to simulation ones?

What additional variables to introduce for this?

What about angular momentum?

Is the key physics actually how material reaches the black hole? Gravitational

torques & viscosity keys?

Berkeley group (Hopkins et al) pursuing this aggressively

 

 

Accreting SPH particles on to the BH

 

wi

wi

wi

Generic feedback physics E=mc2 makes life

easily parameterizable, εr

Factor in efficiency of energy coupling, εf

But is the impact better modelled as heating or momentum?

+How to decide on sphereof influence?

Heating approach (example)  

wi

wi

Note ONB08 apply heating tohalo gas directly!

Momentum approach  

Sphere of influence 4sft

Black hole advection Black hole advection

is trickier than you might think Very important for

accretion calculation N-body integrators

subject to 2-body effects

Want smooth advection Ideally toward

potential well bottom

Black hole advection – SDH05

For low mass BH (<10Mgas)

Find gas part. with lowest PE

Relocate to that position if vrel<0.25 cs

If BH starts to carve void – can get problems

Black hole advection – ONB08

Calculate local stellar density Follows local potential

well Move toward density

maximum Step distance

determined by both velocity and softening limit

Avoids significant 2-body issues

Black hole merger algorithm Can give BH it’s own

smoothing length Or use grav softening

Merge when within certain distance + When grav bound (e.g.

ONB08) Or, when relative

velocity less than circ (e.g. BS09)

Summary of implemented modelsModel Accretio

n modelSPH

accretion

Feedback model

BH advectio

n

BH merger

SDH05 BHL Classic probabilit

y

Heating Lowest local PE

Sound speed

criterionBS09 BHL+alph

a modProb

based on mass

Heating Lowest local PE

Circular vel

criterionDQM11 Viscous

timescaleProb

based on mass limit

Wind Massive tracer

Distance only

ONB08 Drag based

Prob based on

mass

Halo heating

Toward max

density

Grav bound

WT12 BHL Local particles

first

Heating Toward max

density

Sound speed

criterion

Numerical issues Some of these processes involve

very small cross-sections => numerically sensitive

Non-associativity of floating point has an impact Worse in parallel comps –

accumulations come in different orders

We’re still quantifying the impact

Difficult decisions To vary star formation model or not

to vary?

We’ve kept things the same – “classical” model that’s pseudo-multiphase Modified cooling based upon pressure

eqlb between phases Heated regions obvious in plots/movies Can introduce some differences

compared to other researcher’s models (ask me at end)

Simulation models Classic two spiral

merger (very close to Springel et al 2005 model)

End state: red & dead elliptical

Low (~200k particles per galaxy) and mid (~1m) resolution models

Movie 2

SFRs can be numerically sensitive

SFRs are very numerically sensitive, from Springel et al 2005:

Multiphase models suppress passage peak

If the star formation rate is tied togas density, the amplitudes of merger-induced starbursts dependon the compressibility of the gas, which is influencedby both the stiffness of the EOS, as well as dynamic range inresolution of the numerical algorithm.

Results – SFRs

Initial peak fromdisc response

SDH05BS09DQMeDQMONB08WT12

Mid res

Low res

Notice barmode lessstrong

Disk morphology at apoapsis

Movie 3

Results – black hole mass growth

M-σ for mid res final states

ONB08

BS09

DQMe

DQM, SDH05, WT12

Densities & temps “similar”

Results – time stepSDH05BS09ONB08WT12DQMDQMe

Conclusions Very different behaviours – model

assumptions have enormous range Interaction with SF very important

Need to quantify degeneracies between model parameters!

BH tracking is also quite resolution dependent

AGN impact is far harder to model than SF

Thanks for the invite! Acknowledgements:

NSERC Canada Research Chairs Program Canada Foundation for Innovation Nova Scotia Research & Innovation Trust

Observational hope Duty cycle of AGN

activity remains big unknown

Transverse proximity effect (TPE) can measure it

Problems finding enough

background sources

30m class problem?

ForegroundAGN

Backgroundsources

SF & AGN interaction Starburst-AGN connection well known

Obs -> AGN peak activity about 0.5 Gyr after starburst

SF impacts ISM around BH significantly Impacts temperature & accretion rates

How do these factors interplay? Not that well studied in simulations Likely degeneracies between models