DOE/ERy03077-276

Courant Mathematics and

Computing Laboratory

U.S.

Department of Energy

An Asymptotic
J.

Analysis of an

Expanding Detonation
Jones

Research and Development Report

'^iioported
CN (w

by the Applied Mathematical Sciences Drogram of the Office of Energy Research,

o c
vo
.'^
;J

Department of Energy under Contract DE-AC02-76ER03077

t^

"(0

:hematjcs and Computers

o
w o

o en M -H c e o -a
e W W
01 ca

uary 1987

a,
0)

il^s

NEW YORK

UNIVERSITY

UNCLASSIFIED
DOE/ER/03077-276
UC-32 Mathematics and Computers

New York University

Courant Mathematics and Computing Laboratory

AN ASYMPTOTIC ANALYSIS OF AN EXPANDING DETONATION

J.

Jones

January 1987

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

UNCLASSIFIED

-1-

DISCLAIMER
This report was prepared as an account of work .sponsored by an agency of the United States Government. Melther the United States Government nor any aprency thereof, nor any of their employees, makes any warranty, express or Implied, or assumes any lep;al liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that Its use would not infrlnr^e 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 favorlnf^ by the United States Government or any ap;ency thereof. The 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

-11-

An

Asymptotic Analysis of an Expanding Detonation

James Jones
Courant
Institute of

Mathematical Sciences

New York University New York, N.Y. 10012

ABSTRACT

An
is

expanding cylindrically or spherically symmetric detonation

is

analyzed in a regime in which the radius of the detonation

greater than the width of the reaction zone.

much

Under

this

assumption the fundamental equations may be approximated by

a system of autonomous ordinary differential equations for the

flow velocity and a reaction progress variable. The independent

variable of this system
is

a radial variable in the rest frame of

radius of curvature and the detona-

the detonation front.

The

tion speed enter the system as parameters.

this

At zero curvature

system reduces to the plane wave equations of Zeldovich,

von Neumann and Doering. The plane wave equations possess a

degenerate bifurcation point with a nilpotent linear
bifurcates into a saddle
pjirt,

which

node when the radius

is

finite.

Any

smooth transonic solution must pass through the saddle node.

This fact determines the wave speed implicitly as a function of
radius.

To

leading order, the correction to the detonation speed

is

as a function of curvature

proportional to the curvature, on

the basis of formal and numerical considerations.

-111-

An

Asymptotic Analysis of an Expanding Detonation

James Jones
Qjurant
Institute of

12
'

'

Mathematical Sciences

New York University New York, N.Y. 10012

Table of Contents

1.

Introduction

2.
3. 4. 5. 6.

Derivation of the Model

3
11

Reduction of Order for the System (2.9)

Statement of Main Results

18

7.
8.

The Proposed Normal Form The Transonic Critical Point Proof of Theorem 4.1
Conclusion

27
33

References

40 52 53
55

Appendix

1.

Supported

in

pan by

National Science Foundation Grant Nos. MC5-82-07965 and MCS-83-01255.

2.
3.

Supponed

in part by

Army

Research Office, Contraa No. DAAG-29-84-K0130.

Supported in part by U.S. Dept. of Energy,

Contraa No. DE-AC02-76ER03077.

1.

Introduction.

The diverging detonation

is

interesting because the nonis

linear interaction
in a particularly

between the chemistry and the hydrodynamics

displayed
diverg-

sharp form. The theoretical study of detonations

in a

ing geometry began in the late 1940's with the

fey,

work

of Eyring, Powell, Duf-

and Parlin

[9]

and of H. Jones

[12].

These papers were largely con-

cerned with the relationship between the detonation velocity and the finite

diameter of the charge which has come to be called the "diameter effect".

Wood

and Kirkwood (1954)

[19]

were the

first to

elucidate the relationship

between the wave velocity and the radius of curvature of the detonation
wave. This fundamental work has been extended by a number of authors
7, 8, 15].
[3,

In

all

of these works the geometry considered

was

a cylindrically

symmetric charge with the detonation propagating

metry.

parallel to the axis of

in

sym-

An

excellent review of this

problem can be found

is

Fickett and

Davis
ists

[10], pp.

199

229.

The

axial

geometry

useful to both experimental-

and theorists because steady state solutions

exist.

For the experimentalist

this

means

that

stationary

diverging detonations can be produced in the

this

laboratory.

For the theorist

reduces the problem to an analysis of a sys-

tem of ordinary

differential equations.
state, diverging

Although a substantial body of work on steady

tions has accumulated, there has

detona-

been

little

progress in developing an under-

standing of the dynamics of diverging detonations.

One

reason for

this

is

the

presence of nonuniformities that appear in straightforward attempts to per-

form an asymptotic

analysis

of this

problem.

In

the present work,

we

present an analysis of a spherically symmetric detonation, or a radially pro-

pagating

cylindrically

symmetric

detonation

(independent

of

the

axial

coordinate).

Although

this

is

not a steady state problem,

we

will

show

that

to first order in the ratio of reaction zone width to radius of curvature, the

detonation

may be modeled by

a system of

autonomous ordinary

differential

equations in space, with time entering as a parameter.

The nonuniformities

are exhibited as unboundedness of the vector field along the sonic locus, at

which the flow velocity

relative to the
i.e.

shock equals the sound speed.

bounded transonic
through a
critical

solution,

a solution crossing the sonic locus,

must pass

point of the vector field.

We

will

employ the methods of

bifurcation theory to

show

that an appropriate critical point exists.

is

The evo-

lution in time of the detonation

then determined by a shooting problem

critical point.

connecting the state

at the

shock interface to the

Our main
[5, 6], is a

result, derived in conjunction

with the numerical results of Bukiet

demonstration on the level of numerical computation that the leading order

correction to the detonation

wave

velocity

is

proportional to the curvature.

second main result

is

the formal derivation of ordinary differential equa-

tions for the leading order large radius asymptotics for the internal structure

of the expanding detonation wave.

third

main

result

is

the analysis of this

system of ordinary differential equations.

sonic
critical

We

show

that the plane

is

wave

point bifurcates into a unique critical point which

is

a saddle

point, and that this critical point

smooth function of the shock curvature.

normal form
is

is

proposed for the bifurcation point, and a one parameter

unfolding
trait.

presented illustrating the effect of curvature on the phase por-

In a recent work, Bdzil and Stewart

[4]

have independently presented

related asymptotic theory for the dynamics of curvilinear detonations.

They

-3

employ the strong shock approximation and

a specialized rate law for

which

the reaction terminates at a finite distance behind the shock, and coincides

with the sonic transition.

These idealizations permit them to study two

dimensional detonation dynamics and to deal with the characteristic nonuniformities

in

more

traditional

fashion.

Although

Bdzil

and

Stewart

emphasize that their analysis cannot be expected to apply to more general

reaction kinetics,
it

is

interesting to note that they aJso find that the leading

is

order correction to the speed of the detonation wave

curvature.

linear in the

shock

2.

Derivation of the Model. The equations for a cylindrically symmetric,

transport free, reactive, polytropic gas are

m
III

f\

(2.1)

m,

m-

+ P

m~ pr

E,+

mJE -

p)

^(^ ^ P) ^

(P^)<
P

(p^)m _ pR
P''

0.

Here
is

m =

pM

is

the mass flax, p

i

is

the density, u

is

the radial velocity and p

the pressure,

is

the time and

r is

the radial coordinate.

(all

is

the reaaion

progress parameter, which varies from

total

reactant) to

Mail product). The

energy density

is

E =

pe ~

m-

2p, where the specific internal energy e

4-

is

given by

(2.2)

P(7-1)

(1

M^

is

the polytropic gas constant and q

is

the heat released per unit mass by

the complete reaction.

is

the reaction rate function, which

we assume

has

the Arrhenius form

k{\-

p,

^ up, - ypu, -

^^ -

q(y - l)p/?(XT)

X,

+ u\, - R{k,T)
which has
have died

We

desire an asymptotic solution for an outgoing detonation

initialization transients

been propagating for a long time, so that

out.

For such a solution the radius of the detonation must be much greater

than the other relevant physical length scale, the reaction zone width.
reaction zone width

The

w may
some

be defined as the distance from the shock front to

the point at which

fixed fraction

of the reactant has been consumed.

Denote the radius of the shock

tions

as a function of

time by

z{t).

The assump-

on

and

are then

>v

is

and

w
z

where

the

wave

speed. In the
if

- ^

limit the solution

approaches a
the
It
is

plane detonation (or plane shock

the detonation fails).

Thus

wave
well

speed

is

expected to approach the constant plane wave value.

known
final

that the

speed of the plane detonation

is

is

determined by the

initial

and

Hugoniot curves alone and

independent of the reaction rate R and

This means that
z

therefore independent of the reaction zone width.

are linearly related as
r

and

- ^, independently of w.
for a steady plane detonation

The
explicitly

reaction

zone width

may be

defined

by integrating the rate equation

>o

j^

ail.

where/

(0. 1).

The

relevant observation here

is

that

scales as

Let

be the ratio of the reaction zone width

n^^ to a typical

radius.

Since the

wave

speed

is

constant in the plane

i.

wave

limit, :

and

are nearly proportional for

i,)

sufficiently large

If

we assume

that

uq and the plane wave speed

are

0(1), then

and

are

O
V

I
.

We now

transform the differential equations

analysis by translating the radial
t

(2.3) to an appropriate

form for asymptotic

variable

to the rest

frame of the shock and rescaling

and

: to

be 0(1) for

large times.

Define
T

(2.4)

s -

6f

Ut)
X

r(r)

:(t)

= -^^ e

\=i-\x
where
t

= z-\x
that the transformation r

Note

involves a reversal of

ax

orientation, as

shown

in Fig. 2.1.

This convention was chosen so that x will

be positive behind the shock

in the reaction zone.

jc,

The flow

velocity relative
r, r,

to the shock, oriented parallel to

is

v.

Now

change variables from

r,

and u to

T, X, ^,

and v respectively to obtain

(2.5)

ep,

vp,

+ v,p =

~y^j^/^
e^

ev-

vVj

=
Px
P

epx

v/p,

+ 7pv, = ^(7 -

l)p/?(X,r)

- '^^^j ^/^

X,

+ v\, -

/?(x,r).

Figure 2.1

There are two alternate derivations of (2.5) which are worth mentioning.
First,

one could transform to dimensionless variables and

1.

set all

0(1) parame:

ters to

Alternately, rather than taking

that r

w and

x to be (7(1) and

and

large,

one could assume

and

are 0{l) and that u and x are small by

v

rescaling

R and doing an

"inner expansion" in

behind the shock.

The system
{2.5a)

(2.5) has the

form
w^ - h(w, x,
e).

W- + 0(w)

It

will

be seen

in

Section 3 that the desired solution of (2.5) must possess

both subsonic and supersonic regions with a smooth transition between.

such a transition point the quasilinear operator
singularity

At
This

becomes

singular.

may be

interpreted as

turning point of the system.

If

we

attempt to solve for w^ in (2.5a) near the turning point

derivatives blow
cally

we

find that the x

Specifi-

up unless certain
lie in

solvability conditions are satisfied.

h - ew- must

the range of O.

We

will not

attempt here to prove

to the planar (e

asymptotic convergence of solutions of (2.5) as

solution,

0)

but will proceed in the spirit of formal perturbation theory by

exists

assuming on physical grounds that a smooth transonic solution of (2.5)

which

is

differentiable with respect to

e.

The turning

point structure will be

apparent in the final version of the model which

we

will derive in Section 3.

We

thus assume that a representation

p(.x,

t,

=
e),

po(.r, t)

ep-(.r, t)

- PrcJx,

t.

e)

exists in e for p(-x,

f,

where

,.

PremC^'
;

lim

;^

-0
uniformly
in

6pi(x,0
t.

^
also

x and

We

assume analogous representations for

in

\\

\
all

and

i.

Now

expand (2.5) formally

powers of

grouping together

terms of each power. The zeroth order equations are

(2.6) VqPo,

Vo,p

Po

These are the equations for

by Zeldovich, von
tions

one dimensional steady

state detonation studied

Neumann and Doering (ZND). The hydrodynamic

equarate

may be

integrated to obtain p, v, and p as functions of \.

The

equation then constitutes an ordinary differential equation for \ as a function

oi X,
(2.7)

PqVq

- m

constant)

Po ~ Ps

dx

'

Vq
1

where a- =

^7+1
-^

"v

r.

and V

is

the specific volume.

The
The

s subscript indi-

P
first

cate variables evaluated immediately behind the shock.

(2.7) states that the

of equations

mass flux

is

constant in the shock frame.

in the

The second
referred to as
a

equation defines a line of slope

the

-mof

p,

plane, and

is

Rayleigh

line.

The
p
,

third

equations

(2.7)

defines

family

of

Hugoniot curves
with the \

in the

plane.

The

intersections of the Rayleigh line

transitions.

Hugoniot curve determine the possible shock

The

solution terminates at point

where the Rayleigh

is

line

intersects the \

Hugoniot curve.

In general, there

a one parameter family of solutions,

parameterized by the mass flux m, or for a given ambient state ahead of the

10

shock, by the wave speed

^o-

The

first

order equations are

Po(to
PiVo,

(2.8)

po^

VoPi,

v,po,

PoV;,

Vo)

PoVo^

PoVqViv

Po^iVo,,

PiVqVo.,

^ Pu =

Po4o

- vo) - 7Po(to ^(7 - l)(po^i ^ pi^o)

4o

Since the zeroth order equations are steady state, the time derivatives drop

out of the

first

order equations, and

0.
is

Thus the

first

order solution

is

"quasi-steady" in the sense that the solution

given by steady state equations

with time entering as a parameter (through ^o

^d

^q)-

Note also that the tr

in (2.5)

term which appears

in the

denominator of the geometric source terms

order.

does not appear through

of the flow divergence
If

first

Thus,

in the reaction

zone, the deviation

from the value

e.r

at the

shock

is

a second order effect.

the time derivatives and

terms are dropped from (2.5) one obtains a

system of autonomous ordinary differential equations

(2.9)

vp,

+ v,p =

^P(C
C

V)

w, +

11

v\,

- R{\J)

It is

easily verified that (2.9) possesses the

same formal zeroth and

first

order

equations as does the original system (2.5).

On

the basis of the

assumed

regularity of the asymptotic expansion for (2.5)

first

we now

take (2.9) as our

order model for the expanding detonation and turn our attention to an

analysis of this system.

Of course one could

also pursue the

more

usual

course of studying the linear non-autonomous perturbation equations (2.8),

but the autonomous system (2.9) permits a

much

cleaner treatment by the

technique of phase space analysis.

3.

Reduction of Order for the System

in the ratio

(2.9).

Note

in (2.9) that e

and

occur only

so that

is

really a

redundant parameter.
1.

This
scale

redundancy

may be removed by
t

setting

This

inverts
will

the

transformation, so that t
henceforth.

and

z.

These identifications

be made

The steady

state

energy equation

is

The

velocity equation in (2.9) can be written as

\iy\

' vp. -

These two equations may then be added to obtain

12-

Vp

0,

which integrates to yield Bernoulli's Law:

(3.1)

^w- ^

+ Vp ^

fit)

Denote values upstream of the shock by the subscript a

will

(for "ambient").

It

be assumed that the ambient state

is

is

constant and unreacted (X^

0),

and that the ambient flow velocity

zero (u^

0,

or v^

z).

For our

polytropic equation of state, (3.1) becomes

(3.2)

1 _v- +

1
is

c"
1

X.^

2
1

.1

Here

(yp/p)-

the sound speed.

Note

that

we

are able to connect

is

across the shock to the ambient state since Bernoulli's law

one of the

Rankine-Hugoniot jump conditions. This

(3.1).

fact

determines the function /(/) in

Now

eliminate/?, between the velocity and pressure equations to obtain

q{y - [)R
(3.3fl)

v.

=
->

c~

->

V-

^(7 - 1)^(1 - X)exp

c-

A7
V-

iz

^)c'-

By

(3.2) c~

is

known
is

function of v and X.
also a

Therefore the right hand side of

This form of the

the equation above

known

function of v and X.

13

velocity equation

may

be combined with the rate equation

(3.3Z;)

X.

two equations
for u and \.
at

to obtain a self contained system of

For a steady undriven plane detonation, the reaction zone terminates

the

Chapman- Jouguet.

or
is

CJ

point v

c in the

Hugoniot diagram.
Hugoniot curve.

At
This

this
criis

point the Rayleigh line

tangent to the \

terium determines a unique solution of (2.7).

The diverging detonation

weakened by

rarefactions produced behind the shock front and terminates

1

below the sonic point on the weak detonation branch of the X ^

curve.

Hugoniot

This means that a sonic transition must occur in the reaction zone
to the supersonic flow at termina-

from the subsonic flow behind the shock

tion.

However, the denominator

in the velocity

equation (3.3a) vanishes

at a

sonic transition, so for a smooth sonic transition to occur, the numerator

must vanish simultaneously.

mentioned
in

The

sonic transition

is

just the

turning point
is

Section 2,
first

and the condition that the numerator vanish

equivalent through

order to the solvability condition for w,

in (2.5a).

We may
i3Aa)

use substitute v- for c~ in (3.2) to obtain the sonic locus

v2

-^ 7+1
1)/?

^j^f

2y.'-qK. ^

solution of (3.3) which crosses the sonic locus will be called transonic.

We

thus seek a transonic soluuon of (3.3) which satisfies

(3.4/7)

q(y -

- ic-(r -

v)

at

the sonic transition.

solution

(v^.,

XJ

of the system (3.4) will be

14

referred to as a sonic critical point.

yields
/?

Note
If

that

when

x, equation (3.4b)

at
1

the sonic critical point.

the detonation has not failed this

implies X

at

the sonic critical point, consistent with the aforementioned

result that a steady

undriven plane detonation terminates

Jq

at

the

CJ

point.

The wave speed

and

finaJ (X

1)

flow velocity v^y for the plane wave

may be determined from equations

Vcj

(2.7)
at

and (3.4a)

and the condition

ccj^

where ccj
^0

is

the

sound speed
1

the

CJ

point.

The

result

is

(3.5)

[iT - l)qc;- +

(((7.L'2

l)qc;'-

1)"

1)^'=

vcy

2d
y +
1

_L

"'2,12

The

positive solution in (3.5)

must be chosen

in

order to satisfy the jump

conditions at the shock.

The system

will (3.3)

be easier to analyze

if

transformed into a more

conventional form. Define the singular change of variable

exp
(3.6)

Ay
cix'Y-}

dx'

;.

{cix'y - v(x')2)v(.r')

Now
(3.7)

change variables from x to y to obtain

v,

- qiy ^(l

\)k(l

- X)v
-

-^

(z

v)v

-,

^ c- exp

Ax

X^

X)(c-

V-).

Observe that the structure of the phase curves

in the (v, X) plane

is

unal-

tered by this change of independent variable since the transformed vector

15

field

is

proportional to the

initial

vector field.

The

integral curves

have sim-

ply been reparameterized to eliminate the singularity in the denominator of the velocity equation.

The

right

hand side of (3.7)

critical

is

now bounded
is

in the

region of interest, and the sonic

point defined by (3.4)

recognized

as a stationary point of the system.

Several observations about (3.7) can be

first

made immediately.
is

Since the

equation

is

proportional to v, the \ axis

a phase curve of the system.

Likewise, the \
tor of
(0, 1).
1

line

is

a phase curve, since the second equation has a facin a fixed

1

\.
z

These two phase curves intersect

critical

point

When
=
1

is

^, the vector field

stationary.

is

proportional to

X, so that the
is
^c

entire X

line

The

- ^

sonic critical point

thus embedsonic critical

ded

in a
is

manifold of

critical points.

We

will see that the z

i.e.

point

a bifurcation point for the system,

is

a point in the

phase plane

where the topology of the phase curves

the vector field.
section.

unstable to small perturbations of

precise in the following

These notions

will

be

made more

For a fixed ambient

state

ahead of the shock, the Rankine-Hugoniot

jump conditions determine

shock, parameterized by the
define
v^,

one parameter family of solutions behind the

:.

wave speed
and

The

critical

point equations (3.4)

(3.7) will

and

X^-

as functions of z

i.

The desired solution of

connect the ambient state (via the jump conditions) to a sonic

this shooting

critical point;
z

problem defines the functional relationship between

of determining the full

and

z.

The method
clear.

dynamic solution of
z

(3.7)

is

now

We

first

analyze the system allowing

and

to vary as

independent

parameters, thus determining the phase plane structure throughout

some

16

domain

in

parameter space.

We

then

may

recover the leading order dynamics

is

by solving the shooting problem.

first

This paper

primarily concerned with the

half of this procedure,

i.e.

determining the phase plane structure of

[5, 6].

(3.7).

The shooting problem has been solved numerically by Bukiet

After eliminating c^ between (3.2) and the second of equations (3.5),

obtain

we

(3.8)

c^

v2

-qi-^

1)(1

\)

^^(v2

vl),

where
V,

(v^,

+ ^ii' - zi)r~ =
^c

is

the flow velocity at the z

is

sonic bifurcation point.

zq.

With

and

independent, v^

now

a function of z and

We may
first

obtain a single differential equation for v(\) by dividing the

equation in (3.7) by the second to obtain

(3.9)

(1

\){c'-

v-)^ =

^(7 -

1)(1

- X)v -

^~^

c'-

exp

This equation clearly shows the (nonlinear) turning point character of the
equation.
in the

When

^c^

we may

use (3.5) and (3.8) to write equation (3.9)

form
v^
1

2q\ir{\

X)

dv

-r

m dX

The

left

hand side of

this

equation
v: ^

is

the differential of the function

/(v, X)

2flix-(l

- X) ^ + C

17

Denoting by

(\v,

Xj

any fixed reference point, we have

(3.10)

^^-^^^'(^^^)

-'^

^ ~^^'^' - ^-)

The choice
(3.11)

v^

=
(v

v^,

yields the separatrix solution

-

v,)=

- 2q^-iK

I)

for the plane wave. (2.7) forv(X).

This result could also have been obtained by solving

Denote by v - ^(X,

r, i)

the solution of the shooting problem for (3.9)

ciitical point.

connecting the ambient state to a sonic

ical

Evaluating g

at the crit-

point,

one obtains an
v,(z, z)

implicit ordinary differential equation

= g(X,(r,r),z.z)
wave speed
:(t).

for the radius :{t) and the

4.

Statement of Main Results.

The main

result of this paper, taken in

conjunction with Bukiet

is

[5, 6] is that the velocity of the

expanding detonation
is

equal to the plane

wave

velocity plus a correction which

to lowest order

linear in the shock curvature k

tionality
is

1/z.

Moreover the constant of proporThis statement

determined by the separatrix solution of (3.7).

follows on a numerical level from our equations together with their numerical
solution and the comparison to the numerical solution of the full Euler equations by Bukiet.

One consequence

of this result

is

that standard

methods of

computation of detonation waves

the

[13]

which use the experimentaj values of

be

planar

detonation

velocity

can

improved

in

accuracy

by

these

18

corrections.
istry,

Moreover

since the correction can be

computed from the chemfrom some phenomeno-

we

believe that the correction can be predicted

logical equation of state

and rate law.

at least after

the latter have been recali-

brated to reflect the new requirement that they reproduce both planar speeds

and leading order curvature corrections. Such a predictive capability would

minimize the amount of experimental calibration necessary to use
theory
in

this

new

numerical computations.

We
theorem.

begin our analysis of the model (3.7) with the statement of our main

We

will see in

subsequent sections that the linear part of the vector

is

field at the plane

wave

sonic critical point

singular,

and

is

consequently a

bifurcation point for the system.

Theorem

4.1 identifies the effect of curva-

ture in (3.7) as a perturbation of the sonic bifurcation point.

THEOREM
ever
(^mm,
i)

4.1
r,

Assume

that there

is

a radius
a
k^^.^

:,

such that

T^.

< T when-

<

,v

and

<

z.

Then

there

is

>

and

neighborhood

-^max)

of Zq SUCh that
bifurcation point v^

The sonic

(v^y

^JL"(i"

^o))^

",

^b

^^ ^^e

vector field (3.7) bifurcates into a saddle point as k

zero, for alii 6 (imin> ^max)ii)

is

increased from

For K^,^ sufficiently small, and

(inun, ima.x)- ^^e saddle point in

i)

is

the unique sonic critical point of (3.7).

iii)

The

location of the saddle point in the phase plane

: i

is

function of

K 6 (0,
iv)

K^^J and
The

(z^,,,

z^^J.

restriction of the vector field (3.3) to the stable separatrix of the

is

saddle point

continuous, uniformly in

(imm. ^max)-

19-

v)

The unique smooth transonic

solution of (3.3)

is

given by the stable

separatrix of the saddle point.

In [3, 5] Bukiet solved the shooting
tional

problem numerically for the funck.

dependence of the wave speed

on the curvature

The

separatrix
the

solution in

Theorem

4.1

is

computed simultaneously, employing

wave

speed determined by the shooting problem.

indicate that the

The computations by Bukiet

on the curvature
to leading

wave speed depends

linearly

order.

We now
Section 5.

present

some

of the tools which this

analysis will

require.

Those familiar with bifurcation theory for vector

fields

may wish

to skip to

By the phase flow ^l{yi) of

eral solution of the differential

a vector field

dy

f(w)

we mean

the geninitial

equation interpreted as a

map

taking an

point

at v

to

its

image

at

some

later value of v.

Thus

^<t,[(w) = f(w) and

(t,^(M)

= w.

DEFINITION
C-'

4.2

Two

vector fields

- f(w) and -y^ = g(w)

that

are

equivalent

if

there exists a

bijection

h(w) such

h(<t):(w))

<}>,nh(w)).

The function h maps

preserving
orientation

the phase flow of

onto the phase flow of g

y.

and parametrization by

C" equivalence

is

also

referred to as topological equivalence.

20-

Let

^f

R" be closed with non-empty interior and with BK smooth.

C-'

We

equip the space of

vector fields on

with the

norm

|fil;. A-

= max
U|

sup
w
^ A'

f(w)

a'w
(/i,

where
3'w

denotes the multiindex

^''w^.
C-'

/),

\i\

i\

'n

7-

and

d'-yii

The norm

||

||^^ provides a precise

measure of the

smallness of a

perturbation of a vector field.

4.3 Let

DEFINITION
tor field
set

^
ay

f(w) be a

vector field in R".

if

The

vec-

g(w)

is

called a
f

C perturbation of size e of f(w)

||(f

there
^

is

a compact

in

R" such that

set of
C-'

= g on R" - ^, and

g)|^ k

<

The
tered

perturbations of size e of
linear space of
field f
C-'

f is just

a ball of radius e cen-

at f in

the

normed
4.4

vector fields.
said to be structurally stable
f
if

DEFINITION
is

A vector

is

there
f.

an

>

such that

all

C^ perturbations of
is

of size e are C^ equivalent to

Structural stability

a precise

form of the notion of robustness of

model.

Many

of the qualitative features of a structurally stable system are

preserved under perturbations of the system.

We

will also

need

local versions of the

above definitions.

DEFINITION
defined

4.5

Two

vector
sets

fields

^
ay

f(w)
in

and
are
a.

-
ay
said

= g(w)
to

respectively
if

on

open

Clf

and

O^
$ ilj

R"

be

equivalent at wq
a

there exists a point w,j

f] Cl^,

C-

homeomor-

phism h(w) and

h(<t>!(w))

neighborhood O'
all

Clj

of

wq such

that

h(wo)

= wq and

4>f(h(w)) for

Q'.

-21

More
point Wq

generally, local equivalence need not be required to preserve a

However, the most common application of

a critical point.
If

local

equivalence

is

in a

neighborhood of
onto a

Any

equivalence must
is

map
its

a critical point

critical point.

the critical point

unique, then

image under an

equivalence

map

is

completely determined.

In this case the

more general

type of equivalence

may be

factored into a translation and a point preserving

equivalence as in Definition 4.5. Similar

Definitions 4.6 and 4.7.

comments may be made concerning

Definitions 4.2 and 4.5 are clearly symmetric with respect to interchange

of f and

g.

They

are also transitive and reflexive (the identity

map

takes f

onto

itself).

DEHMTION
hood

4.6 Let
in
^

^
R".

f(w) be a

vector field in a neighbor-

of

A
Wq
if

vector

field

g(w)
if

is

called

a perturbation of size
hood H'

of

{ at

g(Ho)
||(f

f(wo) and
^,

there

is

a neighbor-

C n

of

Wo

in

which we have

g)||.

<

e.

The CJ perturbations with which we

those obtained by adding to
f

will

be primarily concerned are

the remainder of a Taylor series expansion.

analytic functions

Denote by P^^ the

0(|wp) for
tor field

class of real

on

UC
If

R'"

which are
is

U, where

^
if

is

a neighborhood of the origin.

p,,
z

p(w)

a vec-

on

R'",

and each
,

of the components
U-,

1,

2,

are ele-

ments of
g

P^j^
is

defined on neighborhoods

of the origin, then the field

^ p

'

perturbation of f in a neighborhood

U^
1

q
^
i

U^ of the

origin.

Since the partial derivatives of p vanish at the origin through order;,

f

the size of the perturbation

+ p may be made

arbitrarily small

by

restricting

22

the size of U.

DEFINITION
there
is

4.7

vector field
all

f is

said to be structurally stable at w,,

e

if

an

>
f at

such that
Wq.

C^ perturbations of size

of

f at

w^ are C^
which
f is

equivalent to

By

bifurcation point

we mean

any point

at

not structurally stable.

In general, bifurcation refers to a change in the topological equivalence
class of the vector field

under a perturbation.
,

In practice, the perturbation

is

controlled by a set of bifurcation parameters

classes for the perturbed vector field

and the different equivalence

correspond to different regions of the

parameter space.
point
is

An

n parameter family of perturbations of a bifurcation

called an n
is

parameter unfolding of the bifurcation.

a bifurcation parameter for the plane
is

We

will

show

in

Section 6 that k

wave sonic

bifurca-

tion point, so that (3.7)

a one parameter unfolding of the bifurcation.

Local equivalence
jets [1].

is

usually defined by

means of the elegant machinery of

is

We

have chosen a more prosaic definition which

will also

adequate for our

purposes.

We

make use

of the following restricted

form of

struc-

tural stability.

DEnNITION
tions at Wq.

4.8 Let

K
is

denote a subset of the set of

said to be

all

C^ perturbaat

a vector

field f
A'

C*

stable to class

K perturbations

Wq

if all

perturbations in

of f at

Wq are C* equivalent

to f at Wq.

PROPOSITION
R" with no

4.9 Let f(w) be a continuous vector field in a

domain of

critical points.

Then

f is locally structurally stable.

PROPOSITION

4.10

(Hartman-Grobman Theorem). Let w^ be an

field f(w).
If

iso-

lated critical point of a

smooth vector

none of the eigenvalues

23-

of the Jacobian derivative Df(w^.) of f at at

topologically equivEilent to the linear system

Wq have

a zero real part, then f

w^,) at

is

= Df(w^.)(w -

w^.

Proposition 4.9 follows easily from the Rectification

tinuous vector fields, which states that the vector field
is

Theorem

for con-

C' equivalent to a

constant field in a neighborhood of a noncritical point.

(The Rectification
sufficiently small
all

Theorem may be found,

for example, in
is

Arnold

[2].)

Any

perturbation of a constant field

noncritical and therefore rectifiable, but

nonzero constant
bifurcation point
is

fields

are topologically equivalent.

This implies that a

is

necessarily a critical point.

[1].
It is

proof of Proposition 4.10

presented in Arnold
tion 4.10
is

easily verified that the linear

system

in

Proposi-

structurally stable at the critical point,

and thus the Hartman-

Grobman theorem

implies that a bifurcation point must possess at least one

Critical points possessing

eigenvalue with zero real part.

no eigenvalues with

a zero real part are termed hyperbolic.

The Hartman-Grobman theorem proves

critical

structural stability for hyperbolic

points

by relating them to the

is

critical

points

of

topologically
will

equivalent linear system which

easy to analyze.
point
for
is

The method we

employ

to

analyse the sonic bifurcation

analogous to what the

critical

Hartman-Grobman theorem accomplishes

hyperbolic

points,

except that nonlinear terms are required because of the degeneracy of the
bifurcation point.

We

transform the vector

field in a

neighborhood of the

bifurcation point by a

smooth change of variables

form.
It

to a field

which

is

topologi-

cally equivalent but simpler in

turns out that in most cases, the

topological equivalence class of a vector field at a critical point contains

-24-

polynomial

fields.

In such a case there are fields of smallest degree,

and pos-

sessing the fewest

initiated

number

of terms.

The study of

these minimal fields

was

by Poincare, and the

class.

fields are referred to as

normal forms for the

equivalence

To

find such a field,

we

will take

advantage of the analytiv,

city of the vector field

by expanding

in

powers of the deviations

\ from

the critical values (v^,

1).

We

then seek a diffeomorphism of the phase space

that preserves the linear part of the vector field while eliminating as

many

nonlinear terms as possible.

This procedure

is

customarily carried out order

In
is

by order;

first

quadratic terms are eliminated, then cubic terms and so on.

the present case, only the second degree terms are required. to

The next
finite

step

show

that the transformed field

may be

truncated

at
is

some

order to

obtain a topologically equivalent field.

in part

This latter step

usually the hardest,

because

it

is

usually necessary to identify the class of perturbations

class,

which preserve the equivalence

topological equivalence

and

in part

because the construction of

maps

is

often a highly nontrivial enterprise.

The nor-

mal form thus possesses the same

field,

local topological structure as the original

but

is

much

easier to study.

This method

may

also be applied to an

unfolding of the bifurcation, so

unfolding.

we may speak
first

of a normal form for the

We

will

demonstrate the

step of this technique in Section 6.

As

indicated in the discussion following Definition 4.5, the second step of

is

local topological equivalence

factorized, and the

subproblem of translation
Excellent introductions to

to a fixed critical point

is

also solved in Section 6.

the theory of normal forms for vector fields are available in

Arnold

[1]

and

in

Guckenheimer and Holmes

[II].

25

Assume
lytic

that

is

the matrix of the linear part of a

critical

two dimensional ana-

vector field f(w) with a

point at the origin, so that f has the

form

-^
ay

f(w)

Aw

- f'-i(w)

where

'-^

0(|w|-).

For each integer n ^2, the linear operator

the linear space
Ji, n
ot'

induces a

linear operator

L^ on
/ith

2-vectors having entries which

are

homogeneous

degree polynomials

in w-^,

W2,

(4.1)

(Lvh),

- i l^(Aw), -

(Ah),.

Note

that

L^h
all

is

just the Lie bracket [h,

Aw]

of h with

Aw.

If

L^ were

nonsingular,

nth degree terms in the Taylor series for f could be elim-

inated by the nonlinear change of variables

w = w + L^^h where
In general,

h denotes

the vector of n\h degree terms in the series.

only those nth

degree terms of

which

lie in

the range of

L^ can be

eliminated.

Those

ele-

ments of

J2, n

which do not

lie in

the range of

L^ cannot be eliminated by

a
f

smooth change of variables and are termed resonant. The resonant terms of

contain the essential nonlinear contributions to the phase plane structure.

Applying L^ to the standard basis (w-(w2, 0)^ and

yields a set of vectors

(0, w-fw^)^, j

n,

which span the range of

L;^.

basis for the

range

may

be chosen from
in the set of

this set.

We

need never include more than codim(range(Lx))

This
is

resonant vectors.

because two resonant vectors which

differ only by an

element of range(LA) are smoothly equivalent (the element

identify the

of range(L^)
set
^2.

may be transformed away). Consequently we may

of resonant vectors
/range(L;^).

with the non-zero vectors of the quotient space

basis

Choose any

for

this

quotient space.

These basis

-26

vectors are equivalence classes of elements of

lar

J2. n-

Now

choose any particu-

representatives for these equivalence classes.

set of

These representatives define

into a

maximal

resonant vectors,

i.e.

may be smoothly transformed

vector field with second degree terms consisting of a linear combination of

this

maximal

set of resonant vectors.

In practice this
if

maximal

set is

chosen

from among the standard

basis vectors,

possible, in order to produce the

maximum

simplification of the vector field.

Extensions of these ideas to

higher dimensional vector fields and their associated spaces J

of

homo-

geneous vector valued polynomials are obvious.

In Sections 5

and 6 we begin a

local analysis of the bifurcation point

using the ideas presented above.

In Section 5

we

present and study a one

parameter unfolding of a proposed normal form for the plane wave sonic
critical

point point w

- c,k is

1.

This degenerate

critical

point possesses

two

zero eigenvalues and

the primary bifurcation point for the system.

The

unfolding parameter represents the effect of shock curvature on the phase

plane structure.

For

this simplified vector field the

plane

wave

sonic critical

point bifurcates into a unique saddle point as the shock curvature

Vz

is

increased

from zero. In Section 6 we show

is

that the

unique saddle structure

observed in Section 5

preserved under a class of perturbations which

is

includes (3.7), and that the critical point itself

smooth function of Vz.

5.

The Proposed Normal Form.

V

We propose

-aX
=
V

(5.1)

V (X

tik)v

27

as a

normal form for the vector

field (3.7) at the bifurcation point.

Here

is

the bifurcation parameter.

variables v and
X.

The

coefficients v, a,

and

r\

are posi-

tive.

The

here denote v and X translated to the transonic

at the origin as k is varied.

critical point,

which remains fixed

In this section
results here

we

investigate the properties of this proposed

normal form. The

will assist in

understanding the properties of the transonic

critical point, as

well as lay a foundation for an eventual proof of local topological equivalence

with (3.7)
If

at the critical point.

the

first

of the equations (5.1)

is

divided by the second, a separable

ordinary differential equation

vrfv

= k

dX
r\K

is

obtained for the phase curves v(X).

The general

solution of this equation

is

(5.2)

v:

- 2a X -

Xr

X
TiKln

[X,

T|K

T1K

where

(v^, X^)

denotes any fixed reference point.

For K

>

0, (5.1) has a

unique

critical

point at (0, 0).

The eigenvalues

and corresponding eigenvectors are

p^

v{ar\K)^~,

V. =

:::(TiK/a)^

L'2

Thus the

critical

point

is

a saddle.

-28

The phase plane

horizontal Hne X

structure for (5.1)

tik
is

is

shown

in Fig. 5.1.

Note

that the

a phase curve for (5.1), as well as a horizontal

asymptote for

all

nearby phase curves.

This line corresponds to the

X.

line of the original vector field (3.7) (although the location of the
is

=
is

line

perturbed slightly from t|k).

The region above

the X

t|k line

non-

physical.

Figure 5.1

We may

use (5.2) to eliminate v from the

X^,

equation

in (5.1),

obtaining

sgn(v,)v(X

tik) v;

2a X -

X
Tn<ln

X,

T^K TIK J

-29

The

solution

is

(5.3)

= sgn(v,)J

. 7

v:

2a

i'

Xr

^Kln

^K

J5
J

>r -

T1K

vis

r\K)

When

the vector field (5.1)

becomes

-30

a line of critical points

superimposed over a continuous one parameter family

of phase curves.
(5.5)

Setting k

in (5.2) yields the

phase curve equation

= -(2a)-^v- -

where
^

v;

^ 2aX,.
a family of parabolas, symmetric about the \ axis
0, the phase curve
is

The phase curves are thus

and concave downward.
critical

When P >
where
axis.

a separatrix for the

0, the vertex of the
is

points at (v,., 0),

lies

v^

|P|^"-

When P <
phase curve

parabola

below the v
It is

The p =

tangent to the v

axis at the origin.

this

tangency that produces the second zero eigenstructure of (5.4)

is

value at the origin.

5.2.

The phase plane

illustrated in Figure

Equation (5.5) may be combined with the

an ordinary differential equation
v^

first

of equations (5.4) to obtain

iy2v(v-

- p)

for v(v).

This equation

may be
The

integrated and the solution substituted back

into (5.5) to yield X(y).

results are

-v^tanh(vv^v
-v^coth(vv^.v
(5.6)

/
/

+ +

4>^)
(b+)

p p

> >

[v.l
|v,|

< >

v^
v^

=
1

v^vy

,3 =
I

2
2

,v^tan(vv^,V' /

<!)_)

<0

31

Figure 5.2

cosh(6_)
^

cosh(vv^->' / 2

-r

6_)

> >

|v,|

< >

V,

sinh((b,)
,

|v,|

V,

sinh(vv,.>' / 2 -r

6-)

=
(1

- v,vv/2)cos(6-)

,3 =
,

cos(vv^>'

<

<!>-),

-32-

where

tanh

\-Vr/ vj,

|v,|

<

v,,

coth-i(-v,/v,),

|v,|>v.

We
for

end

this section

with

some observations about

the double zero bifur-

cation point at the origin of (5.4).

There are two choices of resonant terms

in

terms of standard basis vectors. They are

v"

v"

and

v^

v''~^\

Choosing the

latter,

we

find that the only resonant terms in (3.7) at z

^o

are (0, Xv)^, and (0, Xv^)^.

Only the second degree term

Note

is

retained in the

proposed normal form

ing

(5.4).

that the resonant terms (0, v")^ are miss-

from

(3.7) for

all

/i.

As

a consequence of this degeneracy the bifurcation

has

infinite

codimension: the

number

of

distinct

topological

equivalence

classes

which may be obtained by a small perturbation of the plane wave vecis

tor field

infinite.

This exceptional degeneracy

is

exhibited as the line of

in

critical points.

nice presentation of the

nondegenerate case,

which both

second degree resonant terms are present,

may be found

in

Guckenheimer

and Holmes

[11].

The

case of second order degeneracy occurs in models of

33

chemical reactors

[14].

6.

The Transonic

Critical Point.
in

In this section

we

apply the Poincare

transformations described
plane structure of (3.7)
indicated in Section 3

Section 4 to facilitate our study of the phase

in a

neighborhood of the sonic bifurcation point. As

relationship
3,

we

will ignore the functional

in

between

and and

defined by the shooting problem described

Soaion
It

and consider

z
is

z to

be independent parameters of the system.

will

be seen that this

sufficient to

determine the topological structure of the bifurcation point.

In order to carry out the transformations in a

way which preserves

trick of defining

the

correct dependence

on k we employ the standard

an augequation

mented system which consists of (3.7)

together

with

third

0.

We

then expand the augmented system in a Taylor series about the

origin (v, X, k)

(0, 0, 0)

and perform the simplifying nonlinear transforwhile allowing only transformations that

mations as indicated
leave k invariant.
In

in Section 4,

what follows we
It

will use

to

denote a neighborhood of the origin

in

the V, \ plane.

will

be necessary
fO- ^rna-J

at several points to restrict (v, \, k) to

some

cylinder ftmax ^

to obtain a desired result.

For notational
all

simplicity
trictions.

we
Let

will let

k^^^ and 0.^^^ denote the

minimum over

such res-

P denote

the class of real analytic functions on

U C

R'"

which are

C)(|w|") for

w
6.1

U, where

is

any neighborhood of the origin.

PROPOSITION

For each

34

p.,

\2^i
P,,

{1,2},

q,

:,'

{1.2},

Gi, Ut, bi
a,

>

0,

R,;
R,
A:

{3,

,6},

^^ 6

6 {2,

3,4},
<?i

there exist p,

^'3

2,

^2,2' ^n<^

Z?^

R,

/:

6 {2,

3} such that the

system

+ Kpi(v,
Xy
Ky

X, k)

X^i(v, \)

=
=

^^vX + /jtkX

b^K-

- b^X- +

Kp'2(v, X, k)

X^2(^' ^)

is

smoothly equivalent to the system

Vy

(6.1)

= -a^X - bi\\ +

a2'<

+ +

^^^(v, X, k)
Z73K-

X^i(v, X)

ky
Ky
at the origin.

b2K.k

Kp2(v, X, k)

X^2(^' ^)

If

the

augmented system (v

is

expanded
^l-(io-

in a

Taylor series about the point

1,

K,)

{{vh +
system

- z')r\

0),

it

has the
z

form of the
zq,

first

in

Proposition 6.1, providing

<

v^

<

z.

When
plane

this

condition reads
(2.7).
i
is

<

v^y

<

zq.

This

is

satisfied

by the

wave equations
providing

We may
restricted

extend
to

this result

by continuity to more
small

general i

a sufficiently

neighborhood

-35

(-mm- -max) of
fli

zq.

The valucs
(7

of the coefficients

a,,

/?,

for our system are

=
^

\)kq\i,

^2

v^(-'

v*,)

exp

^1

lyk^/,,

^2

- exp

(27(i

- v,)K - A(7 - 1))K)

2A

^3

1731)^ -P
(27(2

- v,)K2 - A(7 -

1)]

v2(4v,

3i)].

The

coefficients ^i, a:,

and

^1 are positive.

The

coefficients bi

and ^3 may

be positive, negative, or zero.

analytically
lytic,
i.

We

point out that the coefficients depend

to (6.1) are ana-

on

{2^,^^ ^ma.v)-

The transformations leading

uniformly

in i 6 {z^^^,

z^^), and thus preserve analytic dependence on

z.

Thus the

p,, q^

terms must depend analytically on

Proposition 6.1

is

proven by a rather lengthy application of the transformation methods outlined

in

Section 4.

These calculations were performed with the

aid

of

Macsyma. The
tion in (6. 1)
is

calculations are

summarized
will

in the

appendix.

The

third equa-

no longer needed and

be discarded.

A phase

plane analysis yields the next result.

6.2
If p.

PROPOSITION
unique
critical

^^

in

(6.1),

the system possesses the

point

-36-

(6.2)

V^

K
2U-y

X.^

= K

a?
fli

At the

critical point, the linear part

Ao(k)

=
a}b^
a-ib-

from

(6.1) has eigenvalues

and eigenvectors given by

(6.3)

aib^K

{(aib^K)2a-,

4a^_biK)^'~

V- =
P=
clear

It is

from

(6.3)

and the signs of the coefficients aj and bi that the

is

critical

point in Proposition 6.2

a saddle for
us that

all i3

R,

i 6 (imin, ^max)-

The Hartman-Grobman Theorem

cally equivalent.
<y,

tells

all

saddle points are topologi(6.1),

Consequently for each fixed k the system

with

p,

and

set to zero,

is

topologically equivalent to the system with b^ also set to

is

zero.

Although the term controlled by b^

resonant and cannot be removed

by a polynomial change of variables, a more general topological equivalence

may indeed remove

tion terms.

this

term.

We

thus group the b^ term with the perturba-

We

next prove the existence of a

small,

critical

point for the system (6.1) for k

to

sufficiently

which

converges

smoothly

the

origin

as

k - 0,

37

uniformly

in

(Proposition 6.3).

Thus

translation of the critical point of

(6.1) to the origin constitutes a

smooth equivalence transformation.

under the perturbation.
<?,

We

also

show

that the saddle point structure persists

PROPOSITION
small
the

6.3 Let p,

^32, and

P2. 2-

Fof
a

*<max sufficiently
critical

system
6

(6.1)

possesses
(i^,,
is

point

(v,(k, :), \,(k, z))

C"([0,

K^^] X

z^^), R-),

corresponding
o^'

through

first

order

in k to (6.2).

There

a neighborhood n^a.x

the origin
in

and a

k^^

such that for

<

<

K^ia^' this critical point is

unique

Ct^^,

uniformly

in z ^ (i^.n' -^max)critical

Proof: The
right

points (v^,

X.^)

of (6.1) are defined by setting the

hand

sides of (6.1) to zero.

By

the Implicit Function

Theorem,
i,

the first
in a
It is

of these equations defines X^ as an analytic function of v^,

and k
0.

neighborhood of the point v

clear

k
it

= 0,z =
is

zo,

provided a^ ^

from

the definition of a^ that

strictly

positive in the region of

interest.

Substituting the expansions

X^

X0(v)

+ kXHv) +

p,
in

=pO(v,X)+
first

Kp|(v, \)

powers of k into the

equation yields

^1

- ^i(v,0)

which

is

well defined for Hj^ax sufficiently small.

Now

substitute this result,

and a similar expansions for p^ into the second

obtain an implicit equation for v^(k):

critical

point equation to

- O(k-) =

0.

The

solutions k

0, v

undetermined eorrespond to the Hne of

Dividing by the
\\-

eritieal

points

discussed in the previous section.

common
lliis is

factor of k,

we

obtain an equation with the solution

point.

at k

=0.

the bifurcation

Observe

that since p,'{\\ 0), ^,(v, 0)

O(v-), the k

equation has

the

form
a^b\

-v.
fli

- 0{\-)

0.

In a sufficiently small neighborhood of the origin, the second factor

is

nonv^,

zero and the v^

is

solution

is

unique.

By

the Implicit Function

Theorem

a single
'

valued smooth function of k in a neighborhood of k =

providing

is

non zero.

This follows from (6.1).

of the origin.

This solution

is

unique

in

some neighborhood

Oj^^.^

Solving for the leading coefficients in

obtain agreement through
first

the Taylor series expansion of v^(k)

in K

we

order

to (6.2).

This result can be substituted back into the expansion for

first

X^(k) to obtain
proof.

order agreement for

\^.

as well.

This completes the

PROPOSITION
small the
critical

6.4 Let p,

?3

and
is

q,

P.. :

For k^^, sufficiently

point in Proposition 6.3

a saddle point.
critical

Proof:
(^c
^c-)!^)
is

We
and

showed

in the

previous proposition that the

point

0(k)

to lowest order.

The perturbation terms

at

p,

are second

order in

v, X,

k, so

when evaluated
system

the critical point they are 0(k~).

The

linear part of the perturbed

at the critical point

thus has the

form

17

Denoting by

(\v,

XJ
vl

any fixed reference point, we have

(3.10)

- 2q^-{\ - A)
V

v^

v^

2c?|X-(l
v^

\,)
.

The choice
(3.11)

\\

=
(v

\f,,

yields the separatrix solution

V,)-

- 2q^-iK

-0

for the plane wave.

This result could also have been obtained by solving

(2.7)forv(\).

Denote by

= g{X,

r, i)

the solution of the shooting

critical point.

problem for (3.9)

at the crit-

connecting the ambient state to a sonic

ical

Evaluating g

point,

one obtains an implicit ordinary differentia] equation

v,(z,i)

= g(X,(z,r),z,i)
wave speed
z{t).

for the radius :(t) and the

4.

Statement of Main Results.

[5, 6] is that

The main

result of this paper, taken in

conjunction with Bukiet

is

the velocity of the expanding detonation

is

equal to the plane

wave

velocity plus a correction which

to lowest order

linear in the shock curvature k

tionality
is

1/z.

Moreover the constant of propor(3.7).

determined by the separatrix solution of

This statement

follows on a numerical level from our equations together with their numerical
solution and the comparison to the numerical solution of the full Euler equations by Bukiet.

One consequence

of this result

is

that standard

methods of

computation of detonation waves

the

[13]

which use the experimental values of

be

planar

detonation

velocity

can

improved

in

accuracy

by

these

18

corrections.
istry,

Moreover

since the correction can be

computed from

the

chem-

we

believe that the correction can be predicted

from some phenomenohave been

recali-

logical equation of state

and rate law.

at least after the latter

brated to reflect the

new requirement

that they reproduce both planar speeds

and leading order curvature corrections. Such a predictive capability would

minimize the amount of experimental calibration necessary to use
theory in numerical computations.
this

new

We
theorem.

begin our analysis of the model (3.7) with the statement of our main

We

will see in

subsequent sections that the linear part of the vector

critical

field at the plane

wave sonic

point

is

singular,

and

is

consequently a

bifurcation point for the system.

Theorem

4.1 identifies the effect of curva-

ture in (3.7) as a perturbation of the sonic bifurcation point.

THEOREM
ever Q
(-^min,

4.1
r,

Assume

that there

is

a radius
a
k^^^^

r;

such that

7 < T when-

<

and
Zq

<

z.

Then

there

is

>

and a neighborhood

4ax) ol

SUCh that
bifurcation point v^

i)

The sonic

(v^y

^."(i"

=^

^o))^

",

^b

i" ^he

vector field (3.7) bifurcates into a saddle point as k

zero, for alii (i^m,
ii)

is

increased from

-^ma.x)-

For

Kniax sufficiently small,

and

(imm' -ma.J' the saddle point

in

i)

is

the unique sonic critical point of (3.7).

iii)

The

location of the saddle point in the phase plane

z i (z^.^,

is

function of

K i (0,
iv)

K^^J and
The

z^^.

restriction of the vector field (3.3) to the stable separatrix of the

is

saddle point

continuous, uniformly in

(-^mm' ^ma.x)-

39

V
(6.5)

(Ao(k)

K-fl(K))

A{k)

where ^(k)
only
if

is

smooth matrix

function.
is

The

critical

point

is

a saddle

if

and

the determinant of
is

A(k)

negative.

This determinant has the form

-Ka2^i + 0{k-), which

position.

negative for k small but positive, proving the pro-

Denote the

critical

point of

(6.1)

by w(k).

Then the

translation

defined by T(w, k)

= w ^

w(k)

is

a global diffeomorphism

mapping

the ori-

gin onto the critical point of (6.1).

that

From
If

Propositions 6.2 and 6.3

we know
by

T depends smoothly on
k),

k.

we denote
g(w, k)

the vector field (6.1)

G(w,
the

and define the translated

relations

field

= G(T(w,

k), k),

we have

commutation

T(t>J^(w), k)

<^(T(w, k))

between the translated and untranslated phase flows.

lowing
result.

We

thus have the fol-

THEOREM

6.5 After translation to the unique critical point in Ci^^^

given by Proposition 6.3, the vector field (6.1) has the form

V
(6.6)

-aX ^
=
V

Kp.,(v, \, k)

^^i(v,
-r

M
Xqii'^',

v\ - T1KV - ookX ^ Kp2(v, X, k)

X)

where
V
r\

bi,

a
co

a^/bi,

a^lay,

abT/b^,

-40-

qi

"2. 2'

^nd p,(w, k) - C>(k|w| + W-).

are identical to those introduced in Section 6.

The

coefficients a. v and
field (6.6) will

r\

The vector

be denoted by g(w, k).

Proof: The
critical

coefficients are obtained by explicit calculation

from the
/?2

point derived in Proposition 6.3.

The conditions on pi and

be

follow

from the requirement

K
[0,Knjax]

that the critical point of (6.6)

at the origin for all

T^J^ provcs the theorem.

7.

Proof of Theorem 4.1 In Section 6 we have identified the phase plane

structure of (3.7) in a neighborhood of the plane

point.

wave

sonic bifurcation

In the present section

we prove Theorem

4.1.

In the course of the

proof

we

will exhibit the topological structure of the solutions for small k in

the physicalh

elevant region of the phase plane.

Define

(7.1)

vl;(v,

X, K, :)

q{y

l)k{l

- X) -

k(z

v)c- exp

Ay

where
(7.2)

c-

is

given by (3.8).
V,

The vector
K,r)

field (3.7)

then becomes

= =

vi|;(v, X,

X,

kil

X)(c-

V-).

As

always,

we assume
id

that q. c^, k, :, k

and A are positive and

that y
0.

>

I.

The function
of possible
Bernoulli's

is

analytic providing

we

avoid a
is

vacuum

state c-

The

set
in

vacuum

states in the v, \ plane

determined by setting

c-

=
is

Law

(3.8),

and

will

be denoted the vacuum locus.

The

result

39

V
(6.5)

(Ao(k) - K-fl(K))

-A(k)

where ^(k)
only
if

is

smooth matrix

function.
is

The

critical

point

is

a saddle

if

and

the determinant of

A(k)

negative.

This determinant has the form

-Kfli^i
position.

+ ^(k-), which

is

negative for k small but positive, proving the pro-

Denote the

critical

point

of
is

(6.1)

by w(k).

Then the

translation

defined by T(w, k)

= w - w(k)

a global diffeomorphism

mapping the

ori-

gin onto the critical point of (6.1).

that

From
If

Propositions 6.2 and 6.3

we know

depends smoothly on

k.

we

denote the vector field (6.1) by

G(w,

k),

and define the translated

field

g(w, k)

= G(T(w,

k), k),

we have

the commutation relations

T(<t,J^(w), k)

<t>KT(w, k))

between the translated and untranslated phase flows.

lowing
result.

We

thus have the fol-

THEOREM

6.5 After translation to the unique critical point in Vl^^^

given by Proposition 6.3, the vector field (6.1) has the form

-a\ (6.6)

Kpi{w, X, k)

-r

Xq-Xw, \)

VX - T1KV - COkX - KP'(V,

X, k)

'^<?2(^'^

^)

where

T|

aj/a^.

a a-/bi, o) ab^bi,

40

q^

i Pi

I,

dSid

p Xyv k) = C)(k|w| + w").

,

The

coefficients a. v and
field (6.6) will

t]

are identical to those introduced in Section 6.

The vector

be denoted by g(w, k).

Proof:
critical

The

coefficients are obtained

by

explicit calculation
/?i

from the

point derived in Proposition 6.3.

The conditions on
be

and p2 follow

from the requirement

K
i

that the critical point of (6.6)

at the origin for all

[0,Kroa.x]

This proves the theorem.

7.

Proof of Theorem 4,1 In Section 6 we have identified the phase plane

structure of (3.7) in a neighborhood of the plane

point.

wave sonic

bifurcation

In the present section

we prove Theorem

4.1.

In the course of the

proof

we

will exhibit the topological structure of the solutions for small k in

the physical!)

elevant region of the phase plane.

Define

(7.1)

>J;(v,

X, K, z)

q(y -

1)A'(1

k)

k(:

v)c- exp

Ay

where
(7.2)

c-

is

given by (3.8).
v,

The vector
K,r)

field (3.7)

then becomes

vil;(v, \,

X,

= kil-

X)ic-

V-).

As

always,

we assume
kj

that q. c^, k. :, k

and A are positive and that 7

>

I.

The function
of possible
Bernoulli's

is

analytic providing

we

avoid a
is

vacuum

state c-

0.

The

set
in

vacuum

states in the v, X plane

determined by setting

c-

=
is

Law

(3.8),

and

will

be denoted the vacuum locus.

The

result

41

2^^JL-(l

- \) =

pL-v-

vl.

The equation
2^ix-(l

- \) ^

V-

^
is

v^
also a parabola.

defining the sonic locus v^

loci

c~
1

The

sonic and

vacuum
axis).

possess the
(3.5)

same vertex
find that v^y

(0,

- v^/2^^,-) and

axis of
1

symmetry (\

From

we

>

2q\i.-,

consequently

v^/2<?^,-

<

for i in
lies

a sufficiently small

neighborhood (imm, imax) of

right intercept with the

^o

so that the vertex

below the v
is

axis.

The

line for the sonic locus

v^

Note

that

From

(3.5)

we have
Zq

2 2

->

\r,

7+1 (7(7
so that
z

1)^

(((7^

- ^)qcr- +

1)'

- \r-cl) >

0,

v^

>

for i sufficiently close to Zq.

This means that the right

branch of the sonic locus up to X

1 lies

to the left of the line \

z.

The

vacuum
c-

locus

is

broader than the sonic locus.

In fact,

from

(3.2)

we

see that

>

is

equivalent to

^(7 - 1)X + cl +
This inequality holds
in a

^^(i-

V-)

>

0.

neighborhood of the rectangle

:

S(i)

{(v, X)

<

<

i,

<

<

1}.

-42

The function
restrict

\\)

is

thus analytic
this

on a neighborhood of S(i), and we

set.

will

now

our attention to

compact
j

The flow
the original

velocity v

of corresponds to a stagnation point u

=
is

in

Newtonian frame of reference.

The

critical

point

{:, 1)

an

artifact of the stagnation point that occurs at the center of the spherical deto-

nation.

Our
This
is

next objective

is

to identify

all

of the critical points of (7.2) in 2(z).

accomplished by considering the zeros of each of the factors of (7.2),

and cataloging the relevant

common

solutions.

When

0, the

line

becomes
value.
3, the
critical
1

a manifold of critical points, each possessing at least

critical points at

one zero eigen-

There are no other

X axis and the
point (v, X)
line

0.

As

pointed out in Section

are fixed phase curves which meet at a fixed

is

(0, 1),

which
identity
0, this

the unique
1,

common

zero of the v and

- X

factors of (7.2).

The

v|;(z,

k, z)

results in a fixed criti-

cal point at (i, 1).

For k

>

is

the unique

common
1) is

zero of the

\\)

and

(1

X) factors.

For positive k, the and the

critical

critical

point (0,

a sink (two nega-

tive eigenvalues),

point

(z, 1) is a

source.

The

factors v

and

c^

v^
is

have no

common

zeros in

21(z).

As shown

in the previous section,

there

a saddle point

on the sonic locus for positive

(v^, 1) as

k,

which converges
is

smoothly to the sonic bifurcation point

at

k -

0.

The saddle point

an intersection of the solution of

il;

with the sonic locus.

We

claim

that for K sufficiently small, these three critical points are the only critical

points of (7.2) in 2(z).

result

It

is

clear that any additional critical point

ij;

must
0, the

from another common solution of

il;(v,

and c-

v~.

At k =

equation

X, k, z)

has the unique solution X

1,

independent of v

-43-

and

i.

More

precisely, for each i.

y\i{vi,
1,

(immJ;

-^max)'
is

^"^ each

v,

[0,

z],

we

have a solution
the
Implicit

0,

i) =

0.
at

Since

smooth on 2(i), we may apply

the

Function Theorem

each of these solutions to show

existence of a unique smooth local solution \

/(v, k, r) in a neighborhood
(7.1),

of (vi, 0, ij) satisfying /(v I, 0, :^)

1.

We

have from

which

is

strictly

nonzero, independent of (and therefore uniformly

in) v^, z;.

This shows that the solution for k

>

is

smooth curve, deformed

slightly

from X =

1.

This

is

the unique solution in a neighborhood

flg

{(v,

\)

<

<

i,

<
\

<

5}

of the \

boundary.

Define S5

S(z)
];

^5.

We

claim that for k suffiSpecifiil;

ciently small,
cally,

no additional branches of

=
there

are created in S(f).

is

we

shall

show

that for every 8

>

k^^ >

such that

is

non-vanishing on the compact set ^^(8, Kj^^x) defined by

^(S,
Fix 8

K^jax)

-^h
1

[0, *<rmx]

[^rain' '"ma..]-

>

0.

Since \

=
By

is

the unique solution of

ib,

ij;

=
a

at

0.

d/

is

nonof

zero on ^"(8, 0)).

A'(8,0) such that

the continuity of

there

is

neighborhood
is

.V5

il;

is

non-zero on
Let
inf

/V5.

Since Kih. 0)

compact we can

choose

A^5 to

be bounded.

x^a.x

is

1^2

\p

q\-

Because K{h. 0)
/r(8,K[jjax)

compact and
\li

.Vg

is

open the infinum

k^^,,).

is

nonzero.

Then

'^6>

so that

is

non-vanishing on K{h.

44

Differentiating (7.1) implicitly with respect to k

we

find

f\

(^

v) c-

A7

which
V

is

negative for v
the
il;

<

and positive for v

>

i,

with a simple zero

=^

at

:.

Thus
and

curve enters S(i) through the v

critical

boundary with
that any

<

1,

exits

through the
i);

point v

1.

Note

two

points in 2(i)

on the

curve are connected by that curve in 2(z).

locus.

The

same
ij;

is

true for

two points on the sonic

For k
once.

sufficiently small, the

curve must cross the sonic locus

in

at least

Now

the slope /v

may
k.

be bounded

an arbitrarily small neighborhood of zero by restricting

is

The

slope of the sonic locus in S(z)

lies

bounded away from

v];

zero, since the

vertex

on the negative X

axis, so that the slopes of the

curve and
z,

the sonic locus in E(z) are in disjoint intervals, uniformly in k and

ficiently small K.

for suf-

Thus the

intersection of the

two curves

is

unique

in S(z).

This

is

just

an example of the general result that transversality of two smooth

is

curves in the plane

stable
(0,

under perturbations of the curves.

l)k)

Note
i|;

that

V\\t

-^(7 -

+ 0(k),

so

that

\\)

>

below the

curve.

Now

define the sectors

{(v, X)

2^

_ as follows.

I-^ =
I._ = 2_. =
I__ The
signs of the

2(z)

il;

> >
<

0, c2

v2
VV-

>

0}

{(v,

X)

2(i) !(:)
I(i)

il;

0, c-

< > <

0}

{(v, X) e

ij;

0, c-

0}

{(v,

X)

ij;

<

0, c-

v=

0}

components of the vector

field (7.2) at

some

point (v, X)

45

(and therefore the quadrant into which the vector points) are determined by
the sector

S-_

containing (v, \), except possibly

at

a critical point or at a

boundary.

We

have seen that the only

ili

critical

point in the closure II__ of

E.^
ij;

is

the saddle point at

c-

v-

0.

phase curve which crosses

=^

away from

a critical point

must do so

vertically, i.e.
at

d\/d\ = n^.

Likewise, a phase curve which crosses the sonic locus

a non-critical point

must have zero slope.

crosses either the
ij;

Further, the sign of the slope of a phase curve which

curve or the sonic locus

at a non-critical point

must
final

change, since exactly one of the components of (7.2) reverse sign.

observation
is

Our
il;

that a non-critical intersection of a

is

phase curve with

or

with the sonic locus in 1(z)

transverse, so that the slope of the phase curve

changes sign
vj;

at

the intersection.
at a

A
ij;

non-transverse intersection with the

curve can only occur

point

where the curve has

a vertical tangent.
in

We

have shown that the slope of the

curve

may be bounded

an arbi-

trarily small

neighborhood of
are possible.

by restricting k^^^, so no non-transverse

intersections

Likewise, non-transverse intersections with the

the sonic locus
is

sonic locus
I^(z).

may be excluded because

never horizontal

in

These considerations make possible a

vj;

classification of the non-critical

crossings with the

curve and with the sonic locus.

Denote by

(v,, \,)

the coordinates of a non-critical crossing.

The

possible crossings of a phase

curve with the

4^-0

curve are

case

<

\,

A.

>

\,

al bi

V, v^

<

0, \, 0, X^

> >

v^

> <

0, \,
0, X,

>
>

>

Vy

-46

cl

V, V,

>
<

0. \,

< <

v,

<

0. \, 0. \,

< <
0.

dl

0, \,

V,

>

The
case

possible non-critical intersections with the sonic locus are

<

V,

>

V,

a2
b2
c2

V, V, V,
V,

> >
< <

0, X,

>

V,

>

0, X, 0, X, 0, X, 0, X,

< > > <

0.

0, Ky
0, X, 0,

<
<

V,
V,

>
<

d2

>

<

Figure 7.1 illustrates cases al-dl.

vl;

al

bl Figure 7.1

cl

dl

We
two

will

show

that the stable separatrix of the saddle point has positive

slope and intersects the X

steps.
First

boundary of S(i).

We

will

accomplish

this in

we show

that the stable eigenspace enters the

2_

sector at

the saddle point, and that the unstable eigenspace docs not.

Then we prove

47

that there are

no subsequent intersections of the separatrix with the

Since

il;

curve or with the sonic locus.

S^.
is

contains no other critical points,

and since the v -

boundary of 2^_

a phase curve, the separatrix must

then exit through the X

boundary.
0, exactly

We

claim that for k

>

one separatrix branch enters each of the

ij;

four sectors.

We

have shown that the

curve

is

transverse to the sonic

locus at the saddle point, so that the sign of the slope of a separatrix which

crosses both curves transversely

is

determined by the sector

^~~

into

which

the corresponding eigenvector points.

into the

Two

such eigenvectors which point

same sector must

necessarily have slopes of the

same

sign.

It

is

therefore sufficient to
locus or to the curve

show
vl;

that the eigenvectors are not tangent to the sonic

0,

and that one of the eigenvectors has positive

slope, and one has negative.

Then one

separatrix will cross with positive

will cross

slope from

2_

into

S_

and the other

from 2_

_ into 5;_ _

with

negative slope.

Let f(v, \) denote the vector field (7.2), and p,

eigenvector
at

an eigenso
that

value

and
==

corresponding

the

saddle

point,

Df V

pV.

By Proposition

6.3, the slopes of

both eigenvectors converge

that

continuously to zero as k - 0, so

we may assume we may

V|

i^

0.

Since

is

only defined modulo a nonzero factor,

set

Vi

1.

Eliminating p

and solving for V;. we obtain

V,

=
:f,
^

The two

solutions correspond to the

two independent eigenvectors.

neces-

sary and sufficient condition for V, to have both a positive and a negative
solution
is fi. 2^2.
1

>

0-

Fo"" (7.2)

we have

-48-

fi.;

= -v^(7 -

O't

+ 0(k) <

0,

and
f2.

= -(1 is

^)(7

+ 0/2 <

0,

so this condition

satisfied.

For

to be tangent to the sonic locus

we must 0'

have

Vf2

0,

0.

After eliminating
is

this

becomes

f|

2^2.

^i. 1^2, 2

or det(Df)

which

impossible
is

at a

hyperbolic
4<

critical point.

The same

result holds

if

the sonic locus

replaced by the

curve, so the separa-

trixes are transverse to

both curves. Thus the slopes of the separatrixes near

the saddle point are given by the sector

points.

S^_

into
is

which the eigenvector

a neighborhood of the

This means that for each fixed k

>

there
lies in

saddle point in which one separatrix branch

each of the four sectors.

_

We
where

next show that the separatrix branch in

2_

connects with the X

boundary of 2(i), with the slope d\/dv =

positive.

f^/fi

of the phase curve every-

We

have demonstrated above that the separatrix leaves the

Since v

saddle point with positive slope.

contains no other critical points,

is

a phase curve, and since

E_

we need

only show that the separatrix does

at

not intersect the

E(z).

il;

=^

curve or the sonic locus

that there
is

a non-critical point in

We
il;

will
0, or

assume

a non-critical intersection of the separatrix

at a contradiction.

with

with the sonic locus, and arrive

Proceedis

ing in the negative v direction

intersection.

from the saddle point

is

into

S_
il;

there
0.

a first

Assume

that the first intersection

with

(The argu-

ment

is

identical for the sonic locus.) This intersection

must have one of the

eliminate

forms al-dl.
cases cl

Since both v^ and \, are positive in

dl.

E__, we may

and

In case bl the phase curve meets the intersection in the

49

positive y direction

from S

Since

we must meet
tiiis

the intersection in the

negative y direction,
case al.

we may

eliminate
il;

case as well, which leaves us with

The slope of the

0.

curve

may

be bounded

in

an arbitrarily
intersect with
is

small neighborhood of
infinite slope,

Since the phase curve \

(b(v)

must

we

see from the definition of case a\ that there

{t)(v)

a neighboril;

hood of the
curve.
(vj, \i)
ij;

intersection in which
ijj

>

vl;(v)

on the 2_

side of the

However,
be a point
at a

is

the upper boundary of

2^

_.

More -

precisely, let
v, intersects

in the interior of

The
where X^

vertical line v

unique point X,
is

|;(V|)

<

Xtits

Since the intersecting

vertical tangent line,

phase curve

smooth,
6(v)

it

is

locally

approximated by

and must

satisfy

<

i|;(v)

locally.

We

have our contradiction.

At

this point

we

will return to a consideration of the vector field (3.3),

x.

with the original independent variable

field possesses the ular, there
is

As

pointed out in Section 3, this

In partic-

same phase curves

as the continuous field (3.7).

a separatrix solution

which increases monotonically from the v

Since the transonic critical point
is

axis to the transonic critical point.

all

unique,

other transonic solutions must cross the sonic locus

points the velocity gradient v\
is

at a non-critical point.
is

At such

unbounded, and the solution

non-smooth.
field (3.3)
is

We

show now

that for fixed k

sufficiently small, the vector

in fact

continuous
exists,

when

restricted to the separatrix, so that a

in i 6

smooth solution w(x)

uniformly

(imm^ ^max)(v^.
,

In Proposition 6.3

we proved
and that
^01

that the transonic critical point

in i in

X^,) is

smooth function of

k,

- X^ = 0{k), uniformly

some neigh-),

borhood (imm' ^max) o^

formly

We

also have the estimate l/jV?!

<

0(k"''

uni-

in z, for the reciprocal of the

eigenvector slope.

The

limit of the ratio

-50-

v^/X^
I/V;.

v^/\^,

as

one approaches the

is

critical

point along a separatrix equals

is

Since X,
if
it

continuous,

this

means

that v^

continuous along the

either side

separatiix

is

bounded

there, since the limits

from

must be

equal.

We

have
lim
v^.

51

asymptotic validity for the model,

increases faster than
x^..

rhis will be so
that this
is

if

the shock radius

We now

show

so.

PROPOSITION
K

7.1

limcc^(K, z) -0

0.

Proof:

We

have
\ (k. :)

jc^(k, i)

<
S

sup
\
:S

|/:~V(\, K, i)exp(A7/c-(\, K, r))|

\

f
^

-j 1

A.

<
In

sup s \ <

|A:--v(X, K,
1

r)exp(Ayc-(X,

k, r))|(-log(l

- \,(k,

i)).

the

proof for Theorem 4.1

in k

we found

that

the

separatrix

solution

is

bounded, uniformly
c

and
is

i,

and bounded away from the vacuum locus

Since
\,-

0,

so the

supremum

bounded.

is

smooth function of k and

z,we have

limKlogfl - \ (K,i)) =

0.

This proves the proposition.

8.

Conclusion. In Theorem 4.1

we show

that

the vector field

(3.7)

possesses the familiar saddle point structure characteristic of

tions.

weak detona-

The

story does not end here, however.

in

The degenerate double zero

eigenvalue bifurcation point present

for

the velocity-reaction progress plane

the

ZND
of

plane

wave
are

is

infinite

codimensionaJ.
to

That

is.

an infinite
topological

number

parameters

required

produce

all

possible

52

equivalence classes which can be obtained by a smooth perturbation of the

system.
It

Only one bifurcation parameter

that a generalization of the

(1/z) arises in the present analysis.

will reveal additional bifurcation"

may be

model

parameters which will produce a more complete unfolding of the bifurcation.

Also, additional bifurcations of the sonic
critical
z

point

may

alter the

phase
to

plane structure

at

values of the shock radius

which are large compared

the reaction zone width but small with respect to the rather stringent limits

imposed

in

Theorem

4.1.

study of these bifurcations could shed light on

phenomena such

as detonation failure

and

reinitiation.

Aknowledgements The
manuscript form in the

results

of

Sections

1-3

were

circulated

in

summer

of 1985 at the Los

Alamos National Labora-

tory and were presented in a public lecture there in January 1986.

J.

We

thank

Bdzil for

comments on these

occasions.

The author

also wishes to thank

B. Bukiet for his efforts in the numerical validation

and implementation of

the model.

critical

reading of a preliminary version of the manuscript by

W.

Fickett,

R.

Menikoff, B. Plohr, and D. H. Sharp provided valuable

insights

and corrections.

The author

is

indebted to

J.

Glimm

for his con-

tinual guidance.

-53

REFERENCES
[1]

V.I. Arnold, Geometrical

Methods

in the

Theory of Ordinary Differential

Equations, Springer Verlag,

[2]

New York

(1983).

V.I.

Arnold, Ordinary Differential Equations, M.I.T. Press, Cambridge

(1978).
[3]
J.

Bdzil and R. Engelke,

"A

study of the steady-state reaction-zone struca heterogeneous explosive," Phys. Fluids

ture of a

homogeneous and

26,

pp. 1210-1221 (1983).

[4]
J.

Bdzil and D. S. Stewart, "The shock dynamics of stable multidimen111.

sional detonation," U.
[5]

Preprint (1986).

B. Bukiet,

"The

effect of curvature

on detonation speed,"

NYU

Preprint

(1986).
[6]

B.

Bukiet,

A Study

of

Some Numerical Methods

for

Two Dimensional

Curved Detonation Problems,

[7]

NYU Thesis

(1986).

E. Dabora,
ble

J.

Nicholls, and R. Morrison,

"The influence of a compressi-

boundary on the propagation of gaseous detonations," Tenth Sympo-

sium (International) on Combustion, pp. 817-830 (1965).

[8]
J.

Erpenbeck, "Steady-state analysis of quasi-one-dimensional reactive

flow," Phys. Fluids, 11, pp. 1352-1370 (1968).

[9]

Eyring, Powell, EXiffey and Parlin, "The stability of detonation," Chem.

Revs. 45, pp. 69-181 (1949).

[10]

W.

Fickctt

and W. Davis, Detonation, University of California Press,

Berkeley (1979).

54

[11] J.

Guckenheimer and

P.

Holmes, Nonlinear
Fields,

Oscillations.

Dynamical S\s-

tems,

and Bifurcations of Vector

Springer V'erlag,

New York

(1983).
[12]

H. Jones,

"A

theory of the dependence of the rate of detonation of soHd

explosives on the diameter of the charge," Proc. Roy. Soc. (London),

A 189,
[13] C. L.

pp. 415-427 (1947).

Mader, Numerical Modeling of Detonations, University of Califor-

nia Press, Berkeley (1979).

[14] L.

Pismen, "Dynamics of lumped chemically reacting systems near singu-

lar bifurcation points,"

Chem. Eng.

Sci.

39, pp. 1063-1077 (1984).

[15]

M.

Sichel,

"A hydrodynamic theory

J. 4, pp.

for the propagation of gaseous deto-

nations,"
[16]

AIAA

264-272 (1966).
effect in

W. Wood and

J.

Kirkwood, "Diameter

condensed explosives,

the relation between velocity and radius of curvature in a detonation

wave,"

7.

Chem. Phys. 22, pp. 1920-1924 (1954).

55-

Appendix.

We

summarize

here

the

construction

of

the

Poincare

transformations leading to equations (6.1).

written in the form
{A. I)

The augmented system may be

Vy

= a^X - aiK - a^v\ + a4KV + a^Kk +

+
Kpi(v, \, k) + X^i(v, X)
biK-X

a(,K^

X.^

bivX +
-

byK-

b^X'

Kpjiv, \, k)

Xq2{v, X)

K,

where v

= v- v^,X=X -

1,/?,

^3

and

<?,

A.

2-

For the augmented

system, the coefficients b2, b^, and b^ are zero, as are the terms kq^ and Kp2,
but
is

we

will

do the more general

case.

The matrix

of the linear part of (A.

1)

(0
(A. 2)

-fli

-a2)

A =

We

seek smooth transformations of the phase space eliminating second order

terms of the vector field while leaving the third variable (here, k) invariant.

The induced
action
(A. 3)

linear operator
basis.

L^ on
The

J3

rnay be computed from (4.1) by

its

on the standard

results are

La

La

f^l

56

VK

57

that {A.\)

becomes
Vy

-a-k - a-K ~
-r

a:?v\

a^KV ~ u^kX
X)

-r

af,K-

Kpi(v, \. k)

A.t?i(v,

\,
K,

= biVk ^ b2K\ - Kp;(v,

\, k)

* X^:(v, \)

=
2

where
tions,

p,

P3

and

^^ 6

A.

2-

Since

we

will

perform several transforma-

we

will hereafter

drop the bars on the coefficients and remainders and

not bother to adopt distinctive notation for the values at each step of the transformation.

Step

2:

Eliminate the ^3 term using the transformation

V
K
V /
V

ai

+
K
/

2a,

The a 4 term
Step
3:

is

replaced by a^

02^3/^ i-

Eliminate the a^ term using the transformation

V A

58

Step

5:

Eliminate the a^ term using the transformation

NOTES

NOTES

NOTES

NOTES

NYU DOE/ER 03077-276 c ? ^'^ Jones, James An asymptotic an expanding analysis of detonatLn

LIBRARY
N.Y.U. Courant Institute of

Mathematical Sciences
251 Mercer
St.

New

York, N. Y.

10012