DOEyER/03077-270

Courant Mathematics and

Computing Laboratory
U.S. Department of Energy

Numerical Analysis and

The
J.

Scientific

Method

Climm and

D. H. Sharp

Research and Development Report

Supported by the Applied Mathematical Sciences subprogram of the Office of Energy Research,
U.S. Department of Energy under

Contract DE-AC02-76ERO 3077

^
<u

Mathematics and Computers

January 1986
x:

4J

R
I

1
-HO
CO

o CO O
Oh
CO

r-

>i4J

-HO)

O Q 3

i-a

i:S m
(0

o E u
3

+J
0)

new YORK

UNIVERSITY

--t

03

UNCLASSIFIED

^, ^. ^ Mathematics and ^ Computers DOE/ER/03077-270 UC-32

Courant Mathematics and Computing Laboratory

New York University

NUMERICAL ANALYSIS AND THE SCIENTIFIC METHOD

J.

Glimm and

D.

H.

Sharp

January 1986

Supported by the Applied Mathematical Sciences subprogram of the Office of Energy Research, U. S. Department of Energy under Contract No. DE-AC02-76ER03077

Los Alamos National Laboratory, Los Alamos, NM

UNCLASSIFIED

-11-

DISCLAIMER
This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any arency thereof, nor any of their employees, makes any warranty, express or Implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply Its endorsement, recommendation, or favoring by The the United States Government or any agency thereof. views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.

Printed in U.S.A.
Available from

National Technical Information Service U.S. Department of Commerce 5285 Port Royal Road Springfield, VA 22l6l

-Ill-

Contents
Page

Abstract
1. 2.
3.

1
2 4

The Computer Revolution The Method of Scientific Computing

The Grand Unified Scheme (GUS)

7 9

4.
5.

Front Tracking

Conclusions
References

12 12
17

6. 7.

Figure Captions
Figures

19

-1-

NUMERICAL ANALYSIS AND THE SCIENTIFIC

METHOD
J.

Glimm

1,2,3

Courant

Institute of

Mathematica] Sciences

New York University New York, N. Y. 10012

D. H. Sharp
Lx)s
*

Alamos National Laboratory

Los Alamos, N. M. 87545

ABSTRACT

The computer has given

rise to a

new mode

of scientific practice,

and today computational science stands beside theory and experiment

as a fundamental tion

methodology.

The impact of

the computer revoluto

on science can be projected from current trends. The demands

will

be made on computing methodologies

be reviewed.
in

One

of the

demands
dologies.

is

an ongoing need for excellence

computational metho-

Generic

difficulties

encountered

in

meeting these challenges will

be discussed. Recent work of the authors and others will be reviewed

in this context.

in pan by the National Science Foundation, grant DMS - 831229. Supported in part by the Applied Mathematical Sciences subprogram of the Office of Energy Research, U. S. Department of Energy, under contract DE-ACX)2-76ERC3077 3, Supported in pan by the Army Research Office, grant D,\AG29-85-01S3 4, Work supponed by the Department of Energy.
1,

Supponed

2,

-2The Computer Revolation.

The ages of
history are demarcated as
fall

1.

much by

the rise and

fall

of ideas and

technology as by the rise and

puter age
is

of empires and nations.

In Jiis sense the

com-

a turning point for

human

events as was the earlier industrial revolution

based upon mechanical energy. With a narrower focus,

the computer revolution

we

consider the impact of

on

science.

Turning points

in science are

marked by

the

introduction of

new

ideas and

new

tools,

and especially by developments which are

calculus, the electron microscope

both tools and ideas.

With the invention of

and

DNA

sequencing, one finds an ability to formulate, understand and solve a range of

this perspective, the

problems which had previously been inaccessible. Judged from

computer revolution
three examples.

will

have an impact on science

at least as large as did these

Very large
In
to

parts of science are described in terms of mathematical equations.

to these equations

most cases the solutions

numerical computation.

cannot be obtained without recourse

In the thirty years since the advent of

main frame com-

puters, an impressive range of problems have been solved with the aid of computers.

Running a modern wind tunnel

to test airfoil designs costs

150 million dollars

per year.
ally

Supercomputers can explore a much larger range of ideas than can actutest facilities

be tested, and expensive

can be reserved for the most promising

designs.

In fact a broad reinge of two dimensional fluid flow problems are under

reasonable scientific control, due primarily to progress on numerical computation on

these problems.

The

list

of such successes could be extended to considerable length.

little

Lx)oking to the future, there can be

subfields of biology
assisted
is

doubt that the mathematization of various

impending.

As

indications

we can

cite the

recent computer

determination of the three dimensional structure of the polio and cold

viruses [37, 49J, computer simulation of the heart with natural and artificial valves
[46]

and algorithms for the analysis of

DNA sequences

[33],

Computers can

also be used as an experimental tool to explore the

in the

unknown,

Feigenbaum discovered universal order

behavior of chaos [20, 17] through

computer experiments, a development which has stimulated and rejuvenated several

-3branchcs of mathematics jmd physics.

Computers can simulate experimentally

difficult

or unattainable parameter

ranges, such as the conditions in the interior of the sun or a few microseconds after
the big

bang origin of the universe. They can also simulate undesirable parameter

ranges such as occur in safety studies to avoid accidents in chemical or nuclear reactors.

There have been three types of

limits

proposed for the ultimate scope of the

computational method.
to be,

However,

it is

not clear

how

firm these limits will turn out

to set ultimate limits
it is

and

in the authors'

jvdgment,

it is

no more possible
in science
It

on

the scope of the computational

methodology

than

to set limits

on the

scope of future progress in theory or experiment.

has been proposed that funda-

mental physical limits such as the speed of

puters.

light will limit the

speed of future com-

However,

the development of parallel computers, which do

many operaclass of

tions concurrently,

may permit an end run around

known
as

this

problem.

com-

binatorial problems,

NP-complete problems, arc

to

effectively outside of
fall into

practical computation.
this class

The extent

which the important problems which

lie in this class

can be restricted to subproblems which do not

is

or can be
it

otherwise modified so as to be effectively computable

not known.

Finally,
in

has

been proposed that problems with producing complex computer code

and reliable form may provide an outer
tain types of

debugged

limit for the use of

computers

to solve certhis

problems.

Again there are several

in particular: portable

strategics

which may mitigate

problem.

We

mention

operating systems and software tools,

high level languages and standardized calling sequences and modular libraries to
allow interchangeability of reliable software components which can be used and thus
tested through time in a variety of applications.

There
scientific

is

no shortage of current

limits.

For

all

of

its

astonishing successes, In

computations provide much

less than science

and technology need.

most important

scientific applications, the solutions arc

undercomputed
is

relative to

the scientific requirements.

stantial

The gap between needs and performance

all

often subthis

(and includes almost

in

three dimensional problems).

Narrowing

gap

will

depend on progress

both hardware and computational methodologies.

To

-4asscss the current achievements and shortfalls of specific computational

like trying to decide

methods
It

is

whether a glass of water

is

half full or half empty.

can

perhaps wryly be observed that those with responsibility for supporting and maintaining a

code tend

to believe that
is

it is

correct and adequate, those

who

use

it

tend

to believe that the

code

rather imperfect, but can nevertheless serve as a guide to

the wise scientist or engineer, and those

who

develop new methods tend to be

acutely

aware of the flaws

in existing codes.

2.

The Method of Scientific Computing.

What

arc the methodologies which define

scientifi';

computing as a

distinct

approach to science, complementing the traditional approaches of theory and experi-

ment?

Scientific computing- begins with

mathematical modeling, whereby essential

features of a scientific problem are expressed in terms of mathematical equations.

Typically, important
to carry out later

compromises and judgments are made

at this stage.

In order

computational steps, unimportant and sometimes important probto a

lem features must be suppressed. These judgments lead

sequence of models

which capture different aspects of the same problem. The different models may be
integrated hierarchically in that a fine scale
ters of a coarser
scientist or

model may be used

to set the

parame-

model, or they may be integrated only

in the final

judgments of the

engineer using them.

After the mathematical equations have been formulated, they must be cast in a

form suitable for numerical computation. This usually requires

tized in

that they be discrc-

one way or another. The key step

is

the formulation of a solution algorithm

is

to solve the discretized equations.

This step

so important that

it

often influences

the preceding steps of

model formulation and

testing

discretization.

The next stage

is

and validation of the solution algorithm. Validation

comparison with known analytic

may be accomplished by
solutions,

the following methods:

comparison with previously validated computations, comprrison with

experiments

laboratory

and internal consistency checks.

Such internal checks

include convergence under

mesh refinement,

ansilysis

of solution errors and analysis

-5of diagnostic data.

Truncation errors and convergence rates can be analyzed

mathematically.

Occasionally there

may be

a mathematical convergence theorem

for the solution algorithm.

The

final stage in scientific

computing

is

to

use a validated and debugged code

for scientific or engineering purposes.

The preceding
as glib as
it

outline for scientific computing

would

strike

most practitioners

omits

all

discussion of difficulties.

We

focus on the difficulties inherent

difficulties defines the subject of

in the solution

algorithm stage.

Dealing with these

numerical analysis.
Typical methodological difficulties, as they manifest themselves to a user,
the following

show

symptoms: Slow convergence, nonconvergence, numerical

effects.
it

instabilities
is

and numerical simulation of spurious physical

Nonconvergence

normally

the sign of an incorrect algorithm, but in listing

tle possibilities,

here,

we have

in

mind more sub-

such as solution convergence in an L norm, while solution values at

specific points (Z,^ convergence)

may be

invalid.

The

trap

is

easy to

fall into,

since

the solution value at

some

singular point (such as a crack tip)

may be

both the most

important and the least convergent part of the solution. The discretization process
is

a modification of the

mathemf

ical

equations and consequcntiy of the physics or

problem formulation which the equations represent. In some cases the original physical

process

is

unstable or only weakly stable to changes in the equations or equaIn such cases, there
is

tion parameters.

the danger that numerical discretization

errors
errors

may

grossly change the nature of the solution.

Even

for stable physics, these

may change

the solution quantitatively to an unacceptable degree.

The symptoms which

cally

a user

may observe

as a sign of

poor performance

typi-

have their origin

in difficulties inherent in the original

mathem-tical equations

and their solutions.

satisfactory

In general finite difference

and

finite

elcmrnt methods give

performance when computing smooth solutions

to regular

problems

in a

space of not too high a dimension.

t)'pes

We
first

want

to

emphasize two somewhat unrelated

of problem difficulties.

The

has to do with rapid solution variation, Isttc

-6space or time derivatives and jump discontinuities or other solution singularities.

Such features are ubiquitous.

They

arise in

and give

rise to

boundary layers.

Chemical reactions are a

systems.

typical source of multiple time scales, also

known

as stiff

Problems with multiple material interfaces or shock waves have discon-

tinuous solutions.

To

explain this point

more

fully

we

note that the shock width in

is

gas dynamics

is

of the order of a

mean

free path,

which

about 10~'

cm and

far

smaller than the macroscopic dimensions of typical flow fields.

dient within the shock might then be of the order of 10* or
units appropriate to the problem.

The pressure

gra-

more using dimensional

Likewise a flame front width could be 10"^

cm

but such numbers are highly problem dependent.

fire flood, the

For example

in

an

oil

reservoir

flame width could be 20

feet,

and

still

small compared to macroscopic

dimensions.
factor of
10^''

Chemical reaction rates within a single problem can easily vary by a

because of their exponential dependence on activation energies. The

importance of discontinuities and singularities has long been recognized by the com-

munity working on hyperbolic equations. However, they are

elliptic

just as

important for

problems such as steady

crack
tips,

state elasticity or multi-fluid incompressible flow,

due

to

corners and material or phase boundaries, a fact which appears to

have been given

less

emphasis than
class of

it

deserves.
difficulties

The second major

problem

we

discuss has to do with the

occurrence of a large or infinite number of essential degrees of freedom. Problems

having these features arise in the study of turbulence,

statistical

mechanics and

equations of state, quantum field theory and stochastic partial differential equations.

This class of difficulties

is

the

more

intractable of the two,

and the examples of

short of the sys-

scientific studies [18, 21, 32, 34, 42, 44]

which can be cited

fall far

tematic needs for computational solutions in these areas due to slow convergence.

Looking ahead

to

our proposed methodology for the

infinite

first class

of problems,

we
also

observe that the problems with an

very
difficult

number of degrees of freedom are

that theoretical ideas

from a theoretical point of view, and

have not

yet provided the

capabilities or

means

to devise numerical

methods with the required enhanced

convergence properties.

-7

3.

The Grand Unified Scheme (GUS).

As

a general principle,

we propose

the

maximal use of

analytic

knowledge of

solution properties.
solutions,

This principle has been successfully used in the czsc of smooth

where

the

known

solution

regularity

is

used to devise higher order

singularities,

methods which give accelerated convergence. For the case of solution

which give
rise to the first class of difficulties

mentioned

in the

previous section,

we

propose maximal use of analytic knowledge of those solution

singularities.

Perhaps

paradoxically, the solution singularities can be a source of problem simplification, in

that the idealized or asymptotic behavior in the

neighborhood of the singularity may

full solution.

possess a simpler structure or higher symmetry than does the

the singularity

Thus

may be an

opportunity for theoretical progress; there

may

also be a
is

considerable body of knowledge concerning the singularities.

this

Our proposal

to use

body of knowledge within the numerical algorithm

knowledge.

itself,

and where necessary

is

to create the required

An

instance of this circle of ideas

the vortex

method, which uses vortices

to represent idealized fluid

elements in (for example)

shear layers and turbulent flow fields [14, 15, 32, 34, 50, 61, 62]. Related ideas arc
the interface

methods based on boundary

integral

methods

[3, 58],

and conformal

mapping

[43].

The Grand Unified Scheme

is

a version of the general principles discussed

above, suitable for compressible reactive fluid flow.

of

These ideas are an outgrowth

work

carried out over several years with

many

colleagues, and

were presented

in

preliminary form in [29].

The scheme could

is

also be called the Kitchen

Smk Scheme

(KSS) because the main idea

to this

to

use everything. There are four main components

briefly in this section.

scheme and we discuss each

The four components

arc:

interior schemes, front tracking, automatic

totic reactive

mesh refinement and automatic asympselection).

built

chemistry (or automatic

mode

High

quality interior

schemes have been

around two ideas: flux limitcrs

to

control overshoots and numerical oscillations and approximate

Riemann

solvers to

achieve upwind differencing

[9, 16,

36, 57].

The main idea of

tional

front tracking

is

to introduce, as

an independent computa-

degree of freedom, the surfaces of jump discontinuity which

may

occur within

a solution. This method will be explained in

more

detail in the following section.

Automatic mesh refinement

solution
is

is

the idea of using fine grids in regions

where the where

singular or rapidly varying, coupled with course grids in regions

the solution has a regular behavior.

enterprise,

This method has given rise to a large ongoing

from which we

cite representative

works

[1, 2, 8, 56J.

Automated asymptotics or automated mode

selection for reactive chemistry

is

the proposal to determine the ambient conditions

from the neighbors of a given

mesh block and on

the basis of these conditions to determine rate limiting reactions

and simplified chemistry, so that the rapid reactions are replaced by frozen or
quasi-steady state conditions.
this

hierarchy of time scales would be identified, and in

hierarchy, the rapid times are treated as quasi-steady state, the intermediate

times would be treated implicitly, while the slowest times would be treated explicitly

and the transients for these time scales would be

fully

and accurately resolved.

this unified

There have been several attempts

In [12] the

to date to

implement

proposal.

modem

interior

schemes were successfully combined with a limited ver-

sion of front tracking.

the above.

In [6], automatic
efforts to

mesh refinement was

successfully

added

to

However,

combine mesh refinement with modern

interior

schemes
at the

in the

absence of tracking led to spurious waves generated by strong shocks

boundaries of the fine meshes. For a problem which does not contain strong

shocks, the successful combination of

mesh refinement with a high

analysis of critical

quality interior

scheme has been obtained

tion asymptotics as

[7].

The automated
is

modes and

solu-

proposed above

an ambitious undertaking and has not been

attempted to date.

Hybrid schemes have obvious software complexity problems which

mentioned below. More serious arc the
to stress the
scientific difficiilties.

will

be
like

Here we would

problems of a fundamental mathematical nature.

Riemann problems

define the scale invariant large time solution asymptotics to leading order and pose a

cumber of problems both

dimensions [28, 30].

for their behavior in the large and their behavior in higher

Such considerations have inspired a study of Ricmann prob-

lems

in a variety of physical contexts

and have led

to the discovery of

new and

strik-

ing mathematical

phenomena

[19, 51, 52, 53, 54], including

shocks which violate

is

most proposed physical admissibility conditions.

bolic

related problem

the hyper-

problem with an embedded

elliptic

region, a notoriously difficult problem

which

arises in a variety of contexts including transonic flow, elasticity, oil reser-

voirs, multi-phase flow

[4, 5,

and the

relativistic

hydrodynamics of a quark-gluon plasma

38, 55, 59, 60].

4,

Front Tracking
Front tracking can be most adequately described by a picture. In Figure 4.1

wc

show

the grids used for a front tracking solution.

Note

that there

is

a regular two

dimensional grid covering a channel, including the wedge, which

the channel.

is

an obstacle in

The front includes a shock wave

incident

on

the

wedge, the channel

represented by a

walls and the wedge, as well as the inlet and outlet surfaces.

It is

lower dimensional (one dimensional) grid, which moves

derived from the jump conditions (the

in time, using velocities

Paemann problem)

of gas dynamics.

The plan

to

use front tracking was initiated by O. McBryan and one of the

authors seven yc2u-s ago; a clear and early statement of the method and the scientific

program
scries of

is

contained

in [22].

This program was developed and implemented in a

the authors, Oliver

more than 40 papers by

McBryan and other coworkers,

see for example [10, 11, 13, 22, 23, 24, 25, 26]. In these papers, proof of principle

and

scientific validation has

been established. Validation has been accomplished by

laboratory experiments, elementary one dimen-

comparison

to analytic solutions,

sional calculations, and previously validated

computer codes,

as well as

convergence

under mesh refinement. These

tests

have been conducted

in the context of a variety

of different applications. In Figure 4.2

we show convergence under mesh

to a simpler

refinement

and comparison of a two dimensional computation

(radially

one dimensional
in

symmetric) computation for a cylindricaUy symmetrical detonation front,

-10-

work by
At

B. Bukict [10]. the outset the concept of front tracking

to initiate this

was considered

to

be hopelessly

wrong, and the plan

controversy.
tifically, it

approach

in a serious

way prompted considerable

scien-

The concerns were of three

to

types: This
it

method might not work

might never be able
to

was too complicated

implement, and

handle

problems of even moderate engineering complexity. Actually there was a record of

previous attempts with this method, which provided encouragement that the method

was fundamentally sound

above.

[40,

45, 48].

The

scientific validation

was discussed

Various numerical issues such as the coupling scheme between the interior
in the

and the front have been examined but deserve further study. Local behavior

neighborhood of co-dimension two intersection points of tracked discontinuities also

needs further study.

The software complexity

ferent

issues

were handled through a strategy which

in

is dif-

from

that normally

employed

computational fluid dynamics, and which

will

might be worthy of consideration by others. These issues have been [27] and
discussed elsewhere.

be

On

the basis of this success with scientific validation and control of software
it

complexity,

can be stated that a proof of scientific principle for front tracking has

been achieved.

Next we discuss the

transition of the front tracking

method from

benchmark validation

studies to problems of

moderate engineering complexity.

is

central research issue involved in this transition

the ability to allow bifurcations or

changes

in front
is

topology due to the interaction or crossing of tracked waves. This

program

partially

complete and

we

report on progress achieved by coworkers.

In Figure 4.3

we show

a computation of a line drive well configuration taken

from an

oil

reservoir study in which the jet pinch off instability leads to successive

front bifurcation [26].

In Figure 4.4
hits a

we show

successive stages in the front topology as a shock

finally a

wave

ramp, forms a regular reflection and

Mach stem

[11].

11-

In Figure 4.5 vvc

ics,

show Kclvin-Helmholtz
slip line exiting

roll

up for compressible gas dynam-

with the tracked

and reappearing through periodic boundaries

[11].

In Figure 4.6

we show

successive stages in front topology as a shock

instability,

wave

hits
J.

a contact in initiation of

Meshkov

taken from unpublished

in
this

work of

Grove
change

[35].

There arc two remarkable points

computation.

One

ij

the

in front

topology which ccurs within the computation, a transition which

has always been regarded as a significant obstacle to the use of this method.

The

other concerns contiguous waves, as discussed in the next paragraph.

There have been several definitive achievements of the front tracking

First

effort.

wc mention
(-y

a calculation of
1.1) SF(,

Grove

[35] in
hit

which a contact separates highly

compressible

from

air.

When

from a shock on the

air side, the

transmitted shock in the

SF(,

and the contact

itself lie

almost on top of each other,

making

this

problem

virtually impossible to

compute correctly by standard methods.

Grove's solution, calculated easily on a 20x20 grid shows the efficacy of front tracking.

J.

Jones [39] determined the leading order endothcrmic effect of radial cooling
in

on reactive chemistry

curved detonation fronts. This piece of asymptotic analysis

it

solved a major open problem in reactive hydrodynamics;

was prompted by

the

requirement of front tracking for analytic knowledge of solution behavior.

King, Lindquist and Reyna [41] solved a 20 year old problem posed by Ratchford
[47],

leading

petroleum engineer.

They showed

that

finite

difference

methods can duplicate the phenomena of viscous fingering with immiscible displace-

ment and

capillary

diffusion,

as

is

observed in laboratory experiments.

This
to

achievement resulted from the need for a benchmark calculation for comparison
front tracking studies [24].

12

5.

Conclaslons.
Science requires computational methods with enhanced capabilities which go

beyond currently available techniques.

needs
is

promising strategy for meeting these

to

to use

methods which are adapted

known

solution behavior.

An

outline

for pursuing this strategy in the context of compressible reactive flow has

been

presented.

Work

directed at implementing this strategy has been discussed with an

emphasis on the front tracking program.

6.

References.

1.

I.

Babushka and
Mechanics.

W.
In:

Rheinboldt. Adaptive Finite Element Processes in StrucElliptic

tural

Problem Solvers n. Eds: G. Birkoff and A.

Schoenstadt, 345-379. Academic Press,

2.
I.

New York

(1984).
Finite

Babushka and W. Rheinboldt. Analysis of Optimal

of

Element Meshes

in

RK Math,
3.

Comp. 33 435-463
I.

(1979).

G.R. Baker, D.

Mciron and S.A. Orszag. Vortex Simulations of the

Rayleigh-Taylor InstabiUty. Phys. Fluids 23 28-64 (1980).

4.

G. Baym. The Hadronixation Transition

J. B. Bell, J.

in the

Early Universe. Preprint.

5.

A. Trangcnstcin and G. R. Shubin. Conservation Laws of Mixed

in

Type Describing Three-Phase Flow

6.

Porous Media. To appear.

M. Berger and

P. Colclla. Local adaptive

mesh refinement

for shock hydro-

dynamics. In preparation.
7.

M. Berger and A. Jameson. Automatic

equations.

adaptive grid refinement for Euler

.

AIAA
J.

Journal 23, 561-568 (1985)

Oliger.
J.

8.

M. Berger and

Adaptive mesh refinement for hyperbolic partial

53, 484-512 (1984).
I. J.

dif-

ferential equations.
9.

Comp. Phys.

J.

Boris and D. Book, Flux corrected transport,

Comp. Phys.

11,

38-69

(1973).

-1310.

B. Bukict. Application of Front Tracking to

tion Fronts.

Two

Dimensional Curved Detona-

NYU preprint.
Glimm,
J.

11.

B. Bukiet, C. L. Gardner, J.

Grove,

J.

Jones, O. McBryan, R.

Menikoff, and D. H. Sharp.

Surface Instabilities and
report.
In:

Applications of Front Tracking to Combustion,

Two

Dimensional Riemann Problems:

conference

Proceedings of the 1985

Army Numerical

Analysis and Compute-

Conference.
12.
I.-L.

Chem

and P. Colella.

A Conservative Front-Tracking

Method

for Hyper-

bolic Conservation
13.
I.-L.

Laws.

LBL preprint.
S.

Chem,

J.

Glimm, O. McBryan, B. Plohr and

J.

Yaniv. Front Tracking for

Gas Dynamics.
14.

Comp.

Phys.

To appear.
Slightly

A. Chorin. Numerical Study of

796 (1973).

Viscous Flows.

J. Fluid

Mech. 57 785-

15.

A. Chorin. Vortex Models and Boundary Layer

1 1-12 (1980).

Instability. J. Sci. Stat.

Comp.

16.

P. Colella
ical

and P. Woodward. The Piecewise Parabolic Method for Gas DynamJ.

Simulations.

Comp. Phys.
Eckmann.

54, 174-201 (1984).

17.

P. Collet

and

J. -P.

Iterated

Maps on

the Interval as

Dynamical Sys-

tems. Birkhauser, Boston (1980).

18.

M.

Creutz. Quarks, Gluons and Lattices. Cambridge University Press,

London

(1983).
19.

Eli Isaacson,

D. Marchesin, B. Plohr and B. Temple. Classification of Solutions

of Quadratic
20.

Riemann Problems.

In preparation.

M. Feigenbaum.

Quantitative universality for a class of nonlinear transforma-

tions. J. Stat. Phys. 91, 25-52 (1978)

and 21, 669-706 (1979).

21.

3.

Gcman and D. Geman.

Stochastic relaxation, Gibbs Distributions, and the

Bayesian restoration of images,

IEEE

trans.

PAMI-6, 721-741 (1984).

-1422.

J.

Glimm,

Eli baacson,

D. Marchcsin, and O. McBryan. Front Tracking for

2, 91-119 (1981).

HypcrboUc Systems, Adv. Math.

23.
J.

Glimm, O. McBryan, R. Mcnikoff and D. H. Sharp. Front Tracking Applied

Rayleigh-Taylor Instability.

to

SIAM J.

Sci. Stat.

Comp. To appear.
S.

24.

J.

Glimm,

Eli Isaacson, B. Lindquist,

O. McBryan and

Yaniv.

Statistical

Fluid Dynamics:
tiers in

The

influence of

Geometry on Surface
I.

Instabilities.

In Fron-

Applied Mathematics, Vol.

The Mathematics of Reservoir Simula-

tion.

Ed. R. Ewing, SIAM, Philadelphia (1984).

Plohr, D. H. Sharp and S. Yaniv.

25.

J.

Glimm, C. Klingenbcrg, O. McBryan, B.

Front Tracking and

Two

Dimensional Riemann Problems. Adv. Appl. Math., 6

259-290 (1985).
26.
J.

Glimm, B.

Lindquist, O.

McBryan and G. Tryggvason. Sharp and Diffuse

In:

Front Tracking and Capillarity.

Proceedings of the

SEG/SIAM/SPE Confer-

ence on Mathematical and Computational Methods in Seismic Exploration and Reservoir Modeling. January 21-24, 1985, Houston, Texas. IP 27.
J.

Glimm

and O. McBryan.

Computational Model for Interfaces. Adv. Appl. Math.

To appear.
28.
J.

Glimm and D. H.

Sharp.

An

S Matrix Theory for Classical Nonlinear Phy-

sics.

Foundations of Physics. To appear.

in

29.

J.

Glimm. Enhanced Resolution

Shock

Wave
-

Calculations.
21,

Presented at the

Argonne National Laboratories,

Mathematics
30.
J.

Sept.

20

Sept.

1984, Computational

State of the

Art Symposium.
Solutions: Their

Glimm. Elementary Waves and Riemann

in Science.

Theory and

their

Role

Proceedings of Seminar on Differential Equations. June 3-7,

1985, Hsinchu, Taiwan.

31.
J.

Glimm. The Role of Better Algorithms.

In: Frontiers of

Supercomputing. U.

California Press.
32.

To

appear.

A. Ghoniem and
Flow:

J. Sethian.

Dynamics of Turbulent

Structiire in

a Recirculating

A Computational Study.

AIAA-85-0146 Submitted

to

AIAA J.

-1533.

W.B. Goad. Computational Analysis of Genetic Sequences. Annual Reviews

Biophysics and Biophysical Chemistry 15 (1986) in press.

in

34.

D. Gottlieb, M. Y. Hussani and

Spectral Methods.

S.

A. Orszag. Theory and Applications of

ICASE

report 83-66,

Dec 1983.

35.

J.

Grove. The Numerical Simulation of Shock Contact Interactions. In prepara-

tion.

36.

S.K. Godunov. Mat. Sbomik 47 271-306 (1959).

37.

J.M. Hogle, M.

Chow

D.J. Filman. Three Dimensional Structure of Polio

Vims
38.

at 2.9

Angstroms Resolution. Science 229, 1358-1365 (1985).

in Elastic Bars.

R.D. James. The Propagation of Phase Boimdaries

Rat. Mech. Anal. 73 125-158 (1980).

Arch.

f.

39. 40.
41.

J.

Jones. Asymptotic Analysis of an Expanding Detonation. In preparation.

R. Lazarus. Private communication.

M. King,
Flow
to

B. Lindquist and L. Reyna. Stability of

Two

Dimensional Immiscible

Viscous Fingering.

SPE

preprint #13953.

42.

S. Kirkpatrick,

CD.

Gelatt and

M.

Vecchi. Optimization by Simulated Anneal-

ing. Science

220 621-680 (1983).

Instability

43.

R. Menikoff and C. Zemach. Rayleigh-Taylor

and Use of Confor-

mal Maps for Ideal Fluid Flow.

44.

J.

Comp. Phys. 51 28-64

(1983).
Teller, E. Teller.
J.

N.A. Metropolis, A.W. Roscnbluth, M.N. Rosenbluth, A.H.

Equation of State Calculations by Fast Computing Machines.
21, 1087-1092 (1953).

Chem. Phys.

45.

G. Moretti.

On

the Matter of Shock Fitting. In: Proc. 4th International Convcrin Fluid

ence on Numerical Methods

Dynamics. R.D. Richtmyer, ed. Springer

Vcrlag, BerUn 287-292 (1975).

46.

C. Peskin and D. McQueen. Modeling Prosthetic Heart Valves for Numerical

Analysis of Blood Flow in the Heart.
J.

Comp. Phys. 37 113-132

(1980).

-1647.

H.H. Ratchford.

Instability in

Water Flooding from Watcr-Wct Porous Media

Containing Connate Water. Trans

48.

ABIE

231 133-148 (1964).

Initial

R. Richtmyer and K. Morton. Difference Methods for

Interscience Publishers,

Value Problems.

New York

(1967).

49.

M. G. Rossmann

et al. Structure of a

Human Cold

Virus and Functional Rela-

tionship to Other Picornaviruses. Nature 317 145-153 (1985).

50.

P.G. Saffman and J.C. Schatzman. Properties of a vortex street of

tices.

finite

vor-

Siam

J. Sci. Statist.

Comput. 2 285-295 (1981).

Bifurcation Theory. Preprint.

51.

M.

Shearer. Shock

Waves and

52.

M.

Shearer and D.G. Schaeffcr. Three Phase Flow in a Porous

Medium and
Proceedings

the Classification of Non-Strictly Hyperbolic Conservation Laws.

of 1985 U.S.
53.

Army

conference on Applied Math, and Computing.

Schaeffer.

M.

Shearer and D.G.

The

Classfication of

2x2 systems of Non-

Strictly

Hyperbolic Conservation Laws, with Application to Oil Recovery.

Comm.
54.

Pure Appl. Math. To appear.

M.

Shearer, D.G. Schaeffer, D. Marchesin and P. Paes-Leme. Solution of the

for a Prototype

Riemann Problem

2x2 System
Mech.

of Non-Strictly Hyperbolic Con-

servation Laws. Submitted to Arch. Rat.

55.

M. Slemrod.

Admissibility Criteria for Propagation of Phase Boundaries in a

f.

van der Waal Fluid. Arch.

56.
J.

Rat. Mech. Anal. 81 301-315 (1983).

Thompson, Z. Warsi and C. W. Mastin. Numerical Grid Generation.

Elsevier, 57.

New York

(1985).
the Ultimate Conservative Difference

B. van Leer,

Towards

Scheme n, HI,

IV, V.

J.

Comp. Phys.

14, 361-370 (1974), 23, 263-275 (1977), 23, 276-299

(1977), 32, 101-136 (1979).

58.

C.P. Verdon, R.L. McCory, R.L. Morse, G.R. Baker, D.I. Meiron and S.A.

Orszag.

Nonlinear Effects of Multifrequency Hydrodynamic

Instabilities

on

Ablatively Accelerated Shells. Phys. Fluids 25 1653-1674 (1982).

-1759.

R. von Miscs. Mathematical Theory of Compressible Fluid Flow, Academic

Press,

New York

(1958).

60.

G. Wallis.
(1969).

One Dimensional Two-phase Flow. McGraw-Hill, New York

61,

N. Zabusky. Contour Dynamics:

A Boundary Integral Evolutionary Method for

Phenomena
S.S.
in Physics

Inviscid Incompressible Flows, In: Nonlinear

ad Biology

Eds R,H. Enns, B.L. Jones, R.M. Miura and

Rangnekar. Plenum Press,

New
62,

York, 125-156 (1981).

N, Zabusky. Contour Dynaics for

2D

Flows. Cont. Math 28, 211-214.

7.

Figure Captions.

Figure 4.1. The grids used by front tracking for a shock on ramp problem.
lar

regu-

two dimensional grid

is

superimposed on a one dimensional front (grid). The

latter consists

of the incident and reflected shock, the

ramp boundary and

the boun-

dary of the computational rectangle.

Figure 4,2. a. Comparison of one and two dimensional computations for a cylindrically

symmetrical detonation front computation,

plot of pressure vs. radius

is

shown. The solid curve shows the

choice computation.

results obtained

by the one-dimensional random

range of pressure values in the

The

vertical lines represent the

two-dimensional front tracking solution at a fixed radius as the angle varies. Thus
the vertical lines

show

the range of angular dependence

on the

solution.

The

grid

is

40 by 40.
Figure 4.2.b Convergence under mesh refinement in the two dimensional computation.

Figure 4.3. Automated bifurcation of tracked fronts illustrated in a

tability in

jet

pinch-off ins-

an

oil

reservoir study.

Here water has been

produced

injected in five wells located

at the

bottom of the figure, and

oil is

at five wells located at the top of the

figure.

The

reservoir parameters represent stable (.^placement, and so the water-oil

interface does not finger.

However

for these

same parameters,

the oil

is

more

-18mobile, and consequently forms a

jet

between the injection wells.

Figure 4.4. Bifurcations of front topology in a shock on ramp problem.

An

incident

shock

hits

a ramp, and undergoes bifurcation to a regular reflection.

When
Mach

the regu-

lar reflection

node reaches the top of the ramp, a bifurcation

to a

type node

occurs. This

Mach

triple point is

degenerate in that the reflected wave has zero

is

strength at the triple point.

The

grid here

30 by 30.
roll

Figure 4.5. Compressible Kelvin-Helmholtz

up, including passage of the front

through a periodic boundary.

Figure 4.6. Initiation of

Mcshkov

instability,

taken from unpublished

work of

J.

Grove. In these four frames, a shock wave

hits

a sinusoidal contact discontinuity

separating two gases of different density (air and SF^).

The

first

frame shows the

wave

fronts just before initial contact, the second after the shock

wave has

hit the

contact, giving rise to transmitted

and reflected shocks

in the

two gases. The

third

and fourth figures show only the main waves, consisting of the contact and reflected and transmitted shocks, after these waves have separated from one another. Contin-

ued

to later time, this

nm

will

show an

interface instability

known

as the

Meshkov

instability.

Figure 4.7. The shock contact interaction, with highly compressible fluid below the
contact, taken

from unpublished work of

J.

Grove.

-19-

'

bo unci dry

S^ctJti\i Shcck

Boui^dar

/?e

iU c t cji

Sli o c /f

Figure 4.1

-20-

2LX

RADIUS

Figure

4. 2.

-21-

20%

10%
/

0%

'il

I
#
I

interior error

froot error

averaie front enor

~**

N-80N-40 N-20
Fig 4.2,b.

N10

N5

^ 2Ax - 2A>

Converge under mesh refinement in the two dimensional computation. Convergence of the front and interior schemes. The pressure errors in the interior and at the front are shown for NzN grids at the time indicated by Fig.^% 2b The # signs represent the interior error, where

Interior Error

= 100% x

lflf\P2.-^Jdxdy '0 '0

,-5.-5 Jo Jo Pimc.I<Uu'^>

The

front error (error bars) gives the range of the errors at the front, defined as

Front Error

= 100% x
[P]

where [P] is the pressure jump at the front in the one-dimensional computation at the same time. The asterisks represent the error of the average pressure behind the front, lamely
P-

Front Error(average pressure)

= 100% x

'^'^ P.~

-22-

Figure 4.3

-23-

time

>

time

0.081

time

0.15

time

22

Figure 4.4

-24-

time

time

0.93

time

2.3

time

2.9

Figure 4.5

-25-

-26-

time

time

" 0.029

time

0.037

time

0.059

Fipure 4.7

FOURTEEN DAYS
A
fine will

be charged for each day the book

is

kept overtiine.