Numerical simulation of compressible

two-phase flows

F. Petitpas
SMASH Team - MARSEILLE

Flows zoology Position of the topic

Compressibles

Single phase

Single velocity

Incompressibles

Multi-phase

Multi-velocity

Interface problems resolution

Non equilibrium two-phase mixtures

Two-phase mixtures in mechanical

equilibrium

General models

Single velocity flows / Interface problems

Evaporation front
Single contact interface

Steam

Media A

Liquid

Media B
Detonation

Some words about discretisation scales

Macro scale : Multi-velocity flow

Micro scale : Single velocity flow

ug

ul

An averaged model with two

velocities is needed

A reduced model for interface

problems is enough
4

On the choice of the method

Sharp interface methods



Lagrangian methods with moving meshes, ALE (Arbitrary

Lagrangian Eulerian)
Good for solid weak
deformation
Not adapted for fluids
computation with extreme
deformations

Front tracking, VOF, Level Set

Very impressive results
Generally non conservative
regarding mass and energy
Heavy numerical treatment

On the choice of the method

Diffuse interface methods

These methods authorize numerical diffusion of interfaces. This presents several

advantages:
 Interfaces are not tracked or reconstructed, they are captured by the numerical
scheme as artificial diffusion zone.
 By the way disappearance or apparition of interfaces are naturally obtained
 Conservative
Volume Fraction of phase 1
1
0.9
0.8

Zone de
mlange =
interface

0.7
0.6
0.5

Liquide
pur

0.4
0.3

Gaz pur
liq <

liq > 1

0.2
0.1
0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

x (m)

On the choice of the model

Euler equations with liquid-vapor Equation of state for evaporation problems

Single phase model with equilibrium EOS (T,p,g,u)
- Able to compute liquid-vapor mixtures at Thermodynamical equilibrium
- But metastable states are omitted
- Unable to treat liquid-gas interfaces

Multi-phase models
4-equation : Euler + mass equation
Thermal and mechanical equilibrium (T,p,u)
- Largely use for gas mixtures where thermal equilibrium condition is not so restrictive
- But unable to treat simple contact interface (interface condition of equal pressure and velocity are
violated)

5-equation : 4 equations + volume fraction equation (Kapila et al., 2001)
Mechanical equilibrium (p,u)
- Able to treat interfaces between non miscible fluids (liquid-gas)
- Able to treat mixture evolving in mechanical equilibrium

6-equation model
Velocity equilibrium (u)

7-equation (Baer & Nunziato, 1986)
Total disequilibrium
- Able to solve a large scale of problems
- Difficult to solve numerically


Outline : The 5-equation model

Topic 1 : Origins and properties of the 5-equation model

Topic 2 : Numerical resolution



The Euler equations

The 5-equation model

Topic 3 : Phase transition with the 5-equation model

Topic 4 : Other extensions





Capillary effects
Compaction effects
Low Mach computing
Etc.

Topic 1 : Origins and properties of the 5-equation model

Starting point : origin of the 5-equation model (The 7-equation model)

Each phase obeys its own thermodynamics (pressure, density, internal energy) and has its own
set of equations :

k


+ .k = (Pk Pk' )
t


kk

+ .(kk u k ) = 0
t

k k u k

+ .( k ( k u k u k + Pk I)) = PI k + (u k ' u k )
t










k k E k

+ .( k ( k E k + Pk )u k ) = PI . k PI (Pk Pk ' ) + .(u k ' u k )
t

= u2

and PI = P1

Z1P2 + Z 2 P1

Z1u1 + Z 2 u 2
P
=
and I
=
Z1 + Z 2
Z1 + Z 2

Baer & Nunziato (1986)

Saurel & al. (2003)
Chinnayya & al (2004)
10

Example of interface problem evolving to a two-velocity mixture

Simulation : Jacques Massoni, SMASH team

11

Asymptotic reduction of the 7-equation model

Why it is interesting to use the 5-equation model instead of the 7-equation

one ?



It is difficult to solve and implies heavy costs regarding CPU and memory.
It contains extra unuseful physics to treat interface problems (two velocities and
two pressures)
Extra physical effects are difficult to introduce (as for example phase transition,
capillary effects, etc.)

Asymptotic reduction by the Chapman-Enskog method :

, = 1
f = f o + f 1

Relaxation parameters tend to infinity

Each flow variable is supporting small variation around an

equilibrium state

12

4 conservative equations

The diffuse interface model (5-equation model)

d1
2c22 1c12 u

= 12

2
2
21c1 + 12c2 x
dt
11 11u
+
=0

x
t
22 22u
+
=0

x
t
u (u + p)
=0
t + x

E + (E + p)u = 0
t
x

+ quation dtat de mlange :

Equilibre mcanique
u1 = u 2 = u
p1 = p 2 = p

Variables de mlange :

= 11 + 22
E = 11E1 + 22 E 2

11p ,1 2 2 p , 2

e
+
1 1
2 1

p = p(, e, k ) =
1

+ 2
1 1 2 1

We will come back on thermodynamic closure in the following

This is a mechanical equilibrium but each phase remains in thermal disequilibrium

13

Physical meaning of the volume fraction equation

k c 2k is the Bulk modulus of media k
It traduces the compressibility of a media
-Big when it is weakly compressible
-Small when it is strongly compressible

u
+u 1 =K
x
t
x

d1
2 c 22 1c12 u

= 1 2

2
2

dt
c
c
1 2 2 x
2 1 1

11 11u
+
=0

x
t
2 2 2 2 u
+
=0

t
x

u (u + p )
=0
t +
x

E + (E + p )u = 0
t
x

u
< 0 1 increases
x

Exemple K < 0
Liquide : phase 1
Bulles gaz : phase 2

u
> 0 1 decreases
x

14

Back to diffuse interface aptitude

Diffusion is due to numerical treatment : Artificial diffusion

By the way, an interface looks like a mixture zone : < k <1
The interface conditions of pressures and velocities equalities are automatically obtained !
Volume Fraction of phase 1
1
0.9
0.8

Zone de mlange =
interface

0.7
0.6
0.5

Liquide pur

0.4

Gaz pur

0.3

liq <

liq > 1

0.2
0.1
0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

x (m)

15

5-equation model properties

Hyperbolic systems : 3 waves speeds.

u cw , u, u + cw

The speed of sound is those of Wood (1930)

Vitesse du son de wood dans un melange eau/air (m/s)
10000

d1
(
)
Kdiv
u
=
dt

11 + div ( u
) = 0
1 1
t



2 2
+ div ( 2 2 u ) = 0

t




u
(
div
u
+

u ) + p = 0
t

E + div [( E + p )u
] = 0
t

1000

1
1
2
=
+
2
2
1c1 2 c 22
c w

100

10
0

0.2

0.4

0.6

0.8

Fraction volumique eau

16

Topic 2 : Numerical resolution considerations

Basics for Euler conservative equations

5-equation model numerical resolution

17

Basics of numerical resolution for the Euler equations

Euler equations:


+ div(u ) = 0
t

Mass balance

Momentum balance

Total energy balance

1D simplification:

u
+ div(u u ) + grad(P) = 0
t


E
+ div((E + P )u ) = 0
t

u
+
=0
t x
u u + P
+
=0
t
x

E u (E + P)
+
=0
t
x

U F
+ =0
t x

U = (, u, E)T
F = (u, u + P, u(E + P))T
18

Computational mesh

i-1

i
i-1/2

i+1

i+2

i+1/2

The cell i is bounded by inlet and outlet sections i-1/2 and i+1/2.
The fluxes cross over these cell boundaries.
The unknowns are computed at the cell center and are piecewise constant functions in
the cell.

f
Example at a
given time

x
How to obtain average cells variables form on time step to an
other one

19

Numerical approximation

i-1

i
i-1/2

i+1

i+2

i+1/2

Integration on cell i over time :

+
div
F
( )dVdt = 0

t V

n +1
i

t *
Fi+1/2 Fi*1/2
=U
x
n
i

The * superscript is for : Solution of the Riemann problem at cell boundary

The Flux F at cell boundaries has been supposed

to be constant during integration time step

CFL condition

20

The Riemann problem

f
fL

At t=0
fR

x0

The initial data are know at a given time and are constant on the right
and left side.
A discontinuity connects the two state.
Question: How does the solution evolves at t>0 ?

21

The Riemann problem solution (advection)

f
f
+u =0
t
x

The two possible solutions are:

x
u
>0
i1/2

In the (x,t) diagram

The solution is:
f (x / t) =

u
<0
i1/2

x/t = ui-1/2

f
if x / t < u
i 1
i 1/ 2
f
i

if x / t > u

i 1/ 2

x
fi-1

fi

22

For the linearized Euler equations

u
+
=0
t x
u u + P
+
=0
x
t

E u (E + P)
+
=0
t
x

The eigenvalues of A are

the waves speeds:

W
W
+ A(W)
=0
t
x

analogue of

f
f
+u =0
t
x

W = (, u, P)T
u

A(W) = 0 u

0
1

Plays the role of a

propagation velocity

+ = u + c , = u c , 0 = u

23

The Riemann problem for linearized Euler equations

t
u-c

WL*

WL

u+c
WR*

WR
x

Solving the Riemann problem consist in determining the perturbed states WL*
and WR* after waves propagation from the known states WL and WR.
24

The Riemann problem solution for linearized Euler equations

t
u-c

u
p*R = p*L
u*R = u*L

pL + LcL u L = p*L + LcL u*L

pL cL L = p*L cL *L

u+c

pR R cR u R = p*R R cR u*R

pR cR R = p*R cR *R
x

From this algebraic set of 6 equations the two intermediate states

WL* and WR* are readily obtained.
25

Riemann problem solution

pL pR + ZRuR + ZLuL
u =
ZR + ZL
*

ZRpL + ZLpR + ZRZL(uL uR )

p =
ZR + ZL
*

*R

p pR
= R +
cR

*L

p pL
= L +
cL

26

P=10 000 bars

P=1 bar

u=0 m/s

u=0 m/s

=10 kg /m3

=1 kg /m3

4500

10

4000

3500

3000

2500

densite

vitesse

Example: shock tube

2000

1500

1000

500

1
0

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Espace (m)

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Espace (m)

Exact solution (lines) / Computed results with Godunov (symbols)

27

Example Supersonic flow around plane profil

28

Summary for Euler equations

The Riemann problem solution is a local solution of the Euler equations between
two discontinuous initial states.
It is the cornerstone of all numerical schemes used in gas dynamics, shallow water
and modern multiphase codes

Recommended literature:
E.F. Toro (1997) Riemann solvers and numerical methods for fluid dynamics.
Springer Verlag

29

4 conservative equations

The diffuse interface model (5-equation model)

d1
2c22 1c12 u

= 12

2
2
21c1 + 12c2 x
dt
11 11u
+
=0

x
t
22 22u
+
=0

x
t
u (u + p)
=0
t + x

E + (E + p)u = 0
t
x

+ quation dtat de mlange :

Equilibre mcanique
u1 = u 2 = u
p1 = p 2 = p

Variables de mlange :

= 11 + 22
E = 11E1 + 22 E 2

11p ,1 2 2 p , 2

e
+
1 1
2 1

p = p(, e, k ) =
1

+ 2
1 1 2 1

We will come back on thermodynamic closure in the following

This is a mechanical equilibrium but each phase remains in thermal disequilibrium

30

Numerical resolutions: issues

1) Volume fraction positivity: How to treat the non-conservative term in the
volume fraction equation when shocks or strong rarefaction waves are
present ?

1


(2c22 1c12 )


+ u.1 = 2
.u
2
1c1 2c2
t
+
1 2
Difficulty to guarantee that

0 < 1 < 1

2) The volume fraction varies across acoustic waves: Riemann solver difficult
to construct.

31

6+1-equation model

Previous difficulties are circumvented using a pressure non-equilibrium model

1

+ u 1 = ( p1 p 2 )
t
x

= 11 + 2 2
E = 11 E 1 + 2 2 E 2

11 11u
+
=0
t
x
2 2 2 2 u
+
=0
t
x
u u 2 + (1p1 + 2 p 2 )
+
=0
t
x

 The pressure equilibrium 5-equation model

is obtained from this 6-equation model in
the asymptotic limit of stiff pressure
relaxation coefficient,
 The speed of sound is monotonic,

1 1 e1 1 1 e1 u
u
+
+ 1 p1
= p I (p1 p 2 )
t
t
x

2 2 e 2 2 2 e 2 u
u
+
+ 2p 2
= p I (p1 p 2 )
t
t
x

6 equations + 1 redundant
equation (coming from the
summation of energies):

( Y1 e1 + Y2 e 2 +
t

c f2 = Y1 c12 + Y2 c 22
 The volume fraction is constant through
right- and left-facing waves when relaxation
effects are absent ( = 0).
1 2

1 2
u ) u (Y1 e1 + Y2 e 2 + u ) + (1p1 + 2 p 2 )
2
=0
2
+
x

32

3-step methods

a)

b)
c)

The (6+1)-equation model is solved without relaxation

effects: Godunov-type scheme,
Stiff pressure relaxation procedure,
Energies reset (in order to ensure energy conservation)
The 5-equation model is solved

33

1st step: Godunov-type scheme

Without relaxation terms, the 6+1

equation model becomes:
1

+u 1 =0
t
x
11 11u
+
=0
t
x
2 2 2 2 u
+
=0
t
x
u u 2 + (1p1 + 2 p 2 )
+
=0
t
x

1 1 e1 1 1 e1 u
u
+
+ 1 p1
=0
t
t
x
22 e 2 22 e 2 u
u
+
+ 2p2
=0
t
t
x

An advection equation
1

+u 1 =0
t
x

A conservative part
U F
+
=0
t x

with

U = (( )1 , ( )2 , u , E )T

F = (( )1 u , ( )1 u , u + p, ( E + p )u )T

A non conservative one

1 1 e1 1 1 e1 u
u
+
+ 1 p1
=0
t
t
x
22 e 2 22 e 2 u
u
+
+ 2p2
=0
t
t
x
34

1st step: Godunov-type scheme

Godunov scheme for advection equation
1

+u 1 =0
t
x

1ni +1 = 1ni

t
( u1 )*i +1/ 2 ( u1 )*i 1/ 2 1ni ( u *i +1/ 2 u *i 1/ 2 )
x

Godunov scheme for conservative equations

U F
+
=0
t x

with

Uni+1 = Uni

t * n n
(F (Ui , Ui+1 ) F* (Uni1, Uni ))
x

U = (( )1 , ( )2 , u , E )T

F = (( )1 u , ( )1 u , u + p, ( E + p )u )T

A non conventional scheme for non conservative internal energies equations

1 1 e1 1 1 e1 u
u
+
+ 1 p1
=0
t
t
x
22 e 2 22 e 2 u
u
+
+ 2p2
=0
t
t
x

(e)kin+1 = (e)kin

t
(eu)k*i+1/ 2 (eu)k*i1/ 2 + (p)kin u*i+1/ 2 u*i1/ 2
x

))

35

Riemann solver
W
W
=0
+ A(W)
t
x

+u 1 =0
t
x
11 11u
+
=0
t
x
2 2 2 2 u
+
=0
t
x
u u 2 + (1p1 + 2 p 2 )
+
=0
t
x

W = (1, s1, s2 , u, p1, p2 )T

1 1 e1 1 1 e1 u
u
+
+ 1 p1
=0
t
t
x
22 e 2 22 e 2 u
u
+
+ 2p2
=0
t
t
x

3 waves speeds

0
0
A(W) = p1 p2

0 0

u 0
0 u

0
0

0 0

0
0
1

u
0

0 0 1c12
0 0 2c22

0
0
2

+ = u + cf , = u cf , 0 = u
with

cf = Y1c12 + Y2c22
2

36

Riemann problem solution

u* =

pL pR + ZR u R + ZL u L
ZR + ZL

ZRpL + ZLpR + ZRZL(uL uR )

p =
ZR + ZL
*

*kL = kL
= kR
*
kR

*
kR

= sk R

with

p = 1 p1 + 2p2
= 11 + 22
Z = c
cf2 = Y1c12 + Y2c22

sk L = s k L
*

Similar to Euler Riemann solver

37

2nd step: Pressure relaxation

1
= ( p1 p 2 )
t

Already solved by the 1st step

1

+u 1 =
t
x

(p1 p 2 )

11e1
= p I ( p1 p 2 )
t

2 2 e 2
= p I ( p1 p 2 )
t
11
=0
t

11 11u
+
=0
t
x
2 2 2 2 u
+
=0
t
x
u u 2 + (1p1 + 2 p 2 )
+
=0
t
x

2 2
=0
t

1 1 e1 1 1 e1 u
u
+
+ 1 p1
=
t
t
x

p I(p1 p 2 )

2 2e 2 22 e 2 u
u
+
+ 2p2
=
t
t
x

p I(p1 p 2 )

u
=0
t
E
=0
t

Relaxation system

38

2nd step: Pressure relaxation

1
= ( p1 p 2 )
t

de1
dv
+ pI 1 = 0
dt
dt

11e1
= p I ( p1 p 2 )
t

e k e 0k + p Ik v k v 0k = 0

de2
dv
+ pI 2 = 0
dt
dt

2 2 e 2
= p I ( p1 p 2 )
t

time integration

1
dv
pIk =
p k dt = 0
0 I
vk vk 0 dt

Y1e1 Y1e10 + Y2e2 Y2e02 + pI1 Y1v1 Y1v10 + pI2 Y2v2 Y2v02 = 0

e e0 + (pI1 pI2 ) Y1v1 Y1v10 = 0

p I1 = p I 2 = p I

Using mass equations

Possible choice

p I = p

Entropy inequality is verified

39

2nd step: Pressure relaxation

Using EOS :

) (

) (

e1 p, v1 e10 p10 , v10 + p v1 v10 = 0

e 2 p, v 2 e 02 p 02 , v 02 + p v 2 v 02 = 0

v1 = v1 (p )
v 2 = v 2 (p )
Closure relation:

1 + 2 = 1 ( )1 v1 (p ) + ( )2 v 2 (p ) = 1

Zero function to solve

Then, we determine:

f (p ) = ( )1 v1 (p ) + ( )2 v 2 (p ) 1
0

p v k (p ) k = ( )k v k

40

3th step: Internal energy reset

We have in the 2nd step determined: p, v k , k

We forget the relaxed pressure but keep volume fractions:

p k

It is then possible to determine mixture pressure by the mixture EOS. By

this way, energy conservation is ensured:
p
p
e 1 1 1 + 2 2 2
2 1
1 1
p new (, e, 1 , 2 ) =
1

+ 2
1 1 2 1
Phasic EOS permits to reset internal energies:
e k = e k (p new , k k , k )

Conservative and good treatment of wave dynamics from both

sides of the interface

41

1D example:
Water-Air shock tube

Peau = 1 bar
eau = 1 kg/m3

Peau = 10 000 bar

eau = 1000 kg/m3

AIR

EAU

Vitesse (m/s)

Pression (Bar)
10000

500

9000

450

8000

400

7000

350

6000

300

5000

250

4000

200

3000

150

2000

100

1000

50

0
0

0.2

0.4

0.6

0.8

0.2

0.4

x (m)

0.6

0.8

x (m)

Densite de melange (kg/m3)

Fraction volumique air

1000

900

0.9

800

0.8

700

0.7

600

0.6

500

0.5

400

0.4

300

0.3

200

0.2

100

0.1

0
0

0.2

0.4

0.6
x (m)

Solution exacte

0.8

0.2

0.4

0.6

0.8

x (m)

Solution numrique

42

Rsultats 2D - Expriences IUSTI (Layes, Jourdan, Houas)

Shock wave
propagation
t = 134s

t = 137s

t = 274s

t = 291s

t = 344s

t = 350s

t = 554s

t = 577s

43

3D example : Missile impact on a metal plate

44

Topic 3 : Phase transition with the 5-equation model

45

Why is the 5-equation model a good candidate for phase transition ?

c2<0

p
P0
sl=cte

s=cte
Liquid
Vapor
Vapor

Liquid

s=cte
sv=cte

Thermodynamic path using the Van de

Waals representation

c2 = v 2

p
<0
v s=cte

Hyperbolicity is lost into the phase diagram

Kinetic path with the 5-equation

model and two separate EOS
Liquid and Vapor isentropes are linked by a kinetic
process
Hyperbolicity is preserved in the entire
46
domain

Phase transition modeling

1) Mass transfer modifies mass equations:

11


+ div (11u ) = m
1 = Y 1
t
2 2


+ div ( 2 2 u ) = m
1 = Y 1
t

dY
d
Avec Y 1 = 1 = 1 1
dt dt

2) Volume fractions change during phase transition:

d1
m
1


= Kdiv (u ) + AQ1 +
dt
I

This interfacial density has

to be determined in order to
close the model.

47

Two-phase model for interface problems with phase transition

11


Entropy analysis for each phase
+ div (11u ) = (g 2 g1 )
t
the mixture
2 2


+ div ( 2 2 u ) = (g 2 g1 )
t


u




+ div (u u ) + ( p ) = 0
t
E


+ div ((E + p )u ) = 0
t
c12 c22
1 2
+
+
2
2
(2c2 1 c1 )
1


1 2
1 2


+ u.( 1 ) =
div( u ) + ( g 2 g1 ) 2
+ H( T2 T1 ) 2
1 c12 2c22
1 c1 2c22
1c1 2c22
t
+
+
+
1
2
1
2
1
2

Mechanical relaxation

Mass transfer

Lois dtat :
e1(1, p) =

p + 1p,1

(1 1)1

e2(2 , p) =

and for

Heat transfer

e = 11e1 + 22e2
+ e0,1

p
+e
(2 1)2 0,2

For the liquid

For vapor

p
p
e 1 1 1 + 2 2 2
2 1
1 1
p(, e, 1, 2 ) =
1

+ 2
1 1 2 1

+ at interfaces (thermodynamical equilibrium)

0 elsewhere (metastable state)

Kinetic parameters : , H =

48

Thermodynamic closure
Assumption : Each fluid is governed by the stiffened gas EOS:

p(, e) =( 1) ( e q ) p
Repulsive effects (gas,
liquids and solids)

Attractive effects
(liquids and solids)

Other useful forms for the SG EOS :

h ( T) = CV T + q

T
s ( p, T) = CV ln
+ q'
1
( p + p )

T
g ( p, T) =( CV q' )T CV T ln
+q
1
( p + p )

For each fluid EOS : 5 parameters to determine :

, P , CV , q, q'
49

Saturation curves for liquid water and vapor water

Psat (Bar)

Lv (kJ/kg)

30

2500

25

2000

20
1500
15
1000
10
500

5
0

300

350

400
T (K)

450

500

300

350

hl (kJ/kg)

400
T (K)

450

500

450

500

hg (kJ/kg)

1200

2800
2700

1000

2600
800

2500

600

2400
2300

400

2200
200
0

2100
300

350

400
T (K)

450

500

2000
300

350

Le Mtayer et al., Int. Journal of Thermal Sciences, 2004

400
T (K)

50

Dodcane liquide

Dodcane vapeur

Pliquide = 1 000 bar

liquide = 500 kg/m3

Pvapeur = 1 bar
vapeur = 2 kg/m3

Pression (bar)

Vitesse (m/s)

1000

350
300

Rarefactions

Phase transfition
front

250

100

200
150
10
100

Choc

50
1
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0
0

0.1

0.2

0.3

0.4

x (m)

0.5

0.6

0.7

0.8

0.9

0.7

0.8

0.9

x (m)

Densite du melange (kg/m3)

Fraction massique de vapeur

1000

1
0.9
0.8
0.7

100

0.6
0.5
0.4
10

Phase transition
front

0.3

Contact discontinuity

0.2
0.1

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

x (m)

Saurel, Petitpas & Abgrall, JFM, 2008

0.9

0.1

0.2

0.3

0.4

0.5

0.6

x (m)

51

High pressure diesel injector

52

High speed motion under water

53

Topic 4 : Some possible extensions

Capillary effects (Perigaud & Saurel, 2005)

Detonation (Petitpas et al. 2009)
Gravity, heat conduction, viscosity, turbulence, etc.

54

Detonation problems

Russian experiences done for the DGA

Air

PMMA

Iron

Explosive

Lif

Explosive+ Al Particles

High velocity
impact for ignition
(1500 m/s)

- 7 fluids
- EOS SG, JWL, IG
- Density ratio : 9000

55

Ebullition crisis simulation

Many physical ingredients are required

g1 = g2
Relaxation

T1 = T2

qcond = T

q
56

Bubble growth

No gravity

Under gravity

57

58