* work supported by the office of naval research. simulation needs overview john petillo – saic,...

* Work supported by the Office of Naval Research.

Simulation Needs OverviewJohn Petillo – SAIC, Burlington, MA

Baruch Levush – NRL, Washington, DC

Sept. 25, 2003

ANL Theory Institute on Production of Bright Electron BeamsArgonne National Lab, Argonne, IL

Sept. 22-26, 2003

Outline

Generate discussion on codes & needs Code needs Needs by component Inter-Code issues Code availability Selection of examples to motivate interest in

needs/wants Action items

Post Processing & Animation

Animation of physics» Provides the researcher with an ability to gain intuition into the dynamics of the

complicated interactions between particles (primary and secondary), fields & the device structure

Electron Gun Multistage Depressed Collector

Code Needs

What’s needed?» Space charge - under what circumstances?» Emission models – characterize , implement model & test» 2D/3D – when is what needed & with what model?» Time-domain

– When can we get away with ES vs. EM?

» Frequency-domain» Need for conformal surfaces – gridding

– Usually means FE, but can be “FD”– FE usually means SLOW – but not necessarily

» What models are needed for modeling efficiency?– Explicit PIC, implicit PIC, transport in smooth pipe, etc.– List these in detail

3D Mesh Types - Illustration

Structured: logical ordering

00

1

12

2

(1,0)

(1,1)

(1,2)

(2,0)

(2,1)

(2,2)

(0,0)

(0,1)

(0,2)

Unstructured: no specific ordering

(0)

(6)

(7)

(1)

(8)

(2)

(3)(5)

(4)(10)

(9)

Needs by component

Examples – what’s needed for…» Guns

– Steady-state/thermionic vs. RF

» Beamline components– Buncher – what’s needed here?

List like this needs to be prepared» Need input from everyone for this

Inter-Code data transfer

Different codes are needed/used in different regions» Incompatibilities may exist – some can be extreme

Need to transfer…» Between codes, often have different…

– Data types– Computational type – Mesh type

» Particle data, field data– Sometimes easy… But, may have to

characterize from one code – process – then send to second code

» Region overlap needed?» How to transfer & start ultra-relativistic beams in a PIC code with a soft

start What development is needed here?

» What standards should be used, or defined?

What Codes Are People Using?

Government provided / home grown / “store bought” These should be listed

» Pros & cons listed

What/which codes can be made available to everyone?

Govt. can commission codes to be written and disseminated to community with training

Training

Proper training often overlooked Introducing new codes to users…

» Many codes are difficult to use – people avoid them– Difficult to use – or difficult to get the right answer? – For difficult codes, often need “Experts” to model first problem before

transitioning code to a user

» Modern codes are both easier and harder to use– Better meshes get better answers– Meshing is now the hard part– Are they faster?

» Outside support from authors is important – must be available

» USPAS tries to help here, but general-purpose codes like PIC codes not well represented in the curriculum

Availability…

Are the codes available, and to whom? Who provides/pays-for codes that cost money? Important Goal…

» that everyone in the community has access to the same tools

» that they are trained

Examples

Motivation for coming up with desired code capabilities

Parallel Plate CapacitorLimited Emission Region

D = 10 cm, R = 10 cm R_beam = 1 cm V = 10,000 V

Is Max right?????

Parallel Plate CapacitorLimited Emission Region

D = 10 cm, R = 10 cm R_beam = 1 cm V = 10,000 V

Parallel Plate CapacitorLimited Emission Region

Childs Law: I = 0.073 A Code: I = 0.23 A – ratio of 3.18 Max says ratio of sqrt(d/r) = sqrt(10) = 3.16!

Example: Use of Conformal Meshes

Guns» Modulation Control grids can be analyzed» Electron surface emission models have improved» Multi-beam and Sheet beam devices can be analyzed

Multistage Depressed Collectors - limitations» Anisotropic Collectors – can be analyzed» Secondary emission models – improved models

Tolerance analysis – often 3D in nature» Most typical alignment and clocking errors can be analyzed

Fine structure representation Multiple particle species – electrons/ions/charge-exchange Multiple-Beam

» No longer a 2D or a small 3D problem when many beams need to be modeled

3D Collector

Multi-Beam Gun

Gridded Gun

Hybrid Multiblock Mesh Example

Building a Hybrid Multiblock Mesh

1. Build structured mesh with an interface to unstructured block.

2. Extract quads on interface.

3. Build unstructured mesh.

4. Merge. Create pyramid elements.

+

=>

=

interface

interface 3D Hybrid MeshMICHELLE Result

Models, Meshing & Algorithms

Finite Element Approach – linear, quadratic Grid System Supported - conformal

» Within Voyager GUI with ICEM-CFD mesher– Supports most high-end CAD modelers – we use SolidWorks– Structured Mesh

3D Multi-block, Hexahedral

– Unstructured Mesh 2D - Triangle, Quadrilateral 3D - Tetrahedral, Hexahedral, Prism, Pyramid

– Hybrid Mesh Single run Structured mesh & Unstructured mesh Makes use of compact data storage of a structured mesh for

computational efficiency

» Within STAR’s ANALYST code– Adaptive Mesh Refinement– Unstructured Mesh

2D – Triangle 3D - Tetrahedral

HexahedraTetrahedra

Coarse Mesh Fine Mesh

Example: Advanced Models & Algorithms

Particle Tracking – adaptive time step model» Boris push – Structured» Nelson push – Structured, Unstructured

Field Solutions» Self-consistent Electrostatic field solution (CG)» Off-axis B-field expansion» Import externally calculated B-fields (Maxwell 3D)» 2D self-magnetic field model

Emission» Child’s law, Longo-Vaughan Temperature-limited,

Thermionic » Secondary emission » Spent Beam particle launch» Periodic, Reflection» Ion beams / charge exchange

iRtrajectory

startingpoint

crossingpoint

- Applied Voltage +EmitterSurface

Cold beam

Thermal beam

Example: GUI & Post-Processing

Graphical User Interface (GUI)» Problem control – setup and run» New model creation with Setup Wizard» Batch control» Direct mesh visualization» Multiple job queuing» Export of results for post-processing» Embedded Python support for comprehensive

scripting capabilities» Hybrid mesh support» Unification and simplification of translator interfaces

Post Processor» Structure, mesh, fields & particles visualization» Calculation of beam quantities

– Beam profiles.– Collection power,current, efficiency table

& visualization.– Alpha, emittance, velocity spread.– User-defined quantities (via Python).

» Effort– Creation of useful tools to maximize understanding and

design intuition

:Beam Transport issues

Modeling of beam transport in long, thin beamlines often suffers from numerical instability or very slow convergence.

Under some conditions, the beamline can be broken up into n segments, and each segment run separately, patching the solutions together across the interfaces with good results.

» Sufficiently short segments are always stable, and each has faster convergence. In the examples shown with 5 segments, the speed increase was a factor of 2 to 4.

Beam Pipe and Beam: uniform beam not matched to B-field

Split Split Split Split

Beam Pipe: Full & Segmented (5 segments)Full

SegmentedRun 1 Run 2 Run 3 Run 4 Run 5

Start End

Long, Thin Beamlines:Performance Comparison

This case represents beam transport where the beams are very laminar. The segmented method works well in these cases.

Emittance Comparison» Shows overall excellent agreement between single

and segmented cases

Tracking Error Comparison» Shows a maximum of < 2.0% tracking difference at

end of trajectory between single and segmented cases

Indications that segmentation is a suitable solution to a difficult problem over a wide problem class

Automating the segmentation task would be a useful development task

2 10

-74

10-7

6 10

-78

10-7

1 10

-61.

2 10

-61.

4 10

-6

0 0.02 0.04 0.06 0.08 0.1

Emittance of full simulation vs. segmented simulation as a function of axial position

Emittance - Full SimulationEmittance - Segmented Simulation

Em

itta

nce

Axial Position

0.5

11

.52

0 0.0001 0.0002 0.0003 0.0004 0.0005 0.0006

No

rma

lize

d P

erc

en

t E

rro

r in

Ra

diu

s

Radius

CPI XK-8050 Example3D Structured, 2D/3D Unstructured Comparison

Area convergence ratio ~71 – difficult problem Same CAD Model - ICEM-CFD produced meshes "Same” MICHELLE input file

2D Unstructured 3D Structured 3D Unstructured 0.277 A 0.2804 A 0.2852 A» Good agreement with DEMEOS» Expect better agreement when 3D unstructured mesh is refined

3D Structured Mesh 3D Unstructured Mesh2D Unstructured Mesh

Embedding MICHELLE in a Commercial Design Environment

CADCAD

MeshingMeshing

Post-processingPost-processing

MICHELLEMICHELLEMesh

Result

File translationinterface

ConfigSolver setupSolver setup

MICHELLEExisting Design Environment Interface Layer

MICHELLE Standardized mesh, result, and solver configuration file formats enable embedding.

MICHELLE has been embedded in STAR’s Analyst product

Integrated parametric CAD. Automated meshing. Adaptive mesh refinement. Python. Support for field solvers

» Statics» Eigenmodes» Driven frequency

Optimization.

Python-based scripting provides important flexibilityfor problem specific numerical post-processing.

Python-based scripting provides important flexibilityfor problem specific numerical post-processing.

Action Items

Come up with a roadmap for code capability Characterize codes and availability (cost)

» Many codes in other disciplines can be used – list what’s available» Distribute to this group

– List of names & contact info of the group members Characterize what it would take to do a start-to-end simulation

» Gather information from community List development level of desired capability/upgrades

» What do you want?» Some development is important and possibly can be made available right

now» Will provide the basis for the roadmap» Start by e-mailing me code names, your application for that code &

suggested upgrades/needs– jpetillo@bos.saic.com