Effects of salts on metal oxidation and refractory corrosion induced by

molten aluminum alloys

S. Afshar
CIREP-CRNF, NSERC-REFRAL Chair, Ecole Polytechniqe of Montreal,
CRIQ Campus, 8475, Christophe-Colomb road, Montreal (Quebec), H2M 2N9, Canada
C. Allaire
CIREP-CRNF, NSERC-REFRAL Chair, Ecole Polytechniqe of Montreal,
CRIQ Campus, 8475, Christophe-Colomb road, Montreal (Quebec), H2M 2N9, Canada
E. Dajoux
Laboratoire Corrodys
55, rue de Beuzeville, BP9, 50120 Equeurdreville-Hainneville, France

ABSTRACT

For various reasons, molten salts, mainly chlorides and fluorides, are used in the
aluminum industry. In the holding and melting furnaces, these salts may have different
roles such as protecting the metal from oxidation, reducing the metal content of the skim
or removing inclusions or impurities from the metal bath. However, the use of these salts
may also have some undesirable effects such as easy attack of the furnace lining
refractory materials by molten aluminum.
The aim of the present investigation was to clarify and to compare the effects of
chloride versus fluoride salts, on the corrosion process of the aluminosilicate based
refractory materials. A laboratory test and procedure, simulating the conditions at the
metal line in the aluminum treatment furnaces, was used. Different aluminum alloys (Al5wt.%Mg, Al-5.5wt.%Zn and 7075 alloy) and chloride or fluoride salts (MgCl2, KCl,
CaCl2, NaCl, MgF2, CaF2, NaF and cryolite) were used as well as several aluminosilicate
materials. Thermogravimetric measurements were also carried out in order to analyze the
effect of the salts on the oxidation of molten aluminum alloys. The experimental results
showed that fluoride salts are more aggressive than chlorides against the tested refractory
products. It is postulated that the fluorides prevent the formation of a protective oxide
layer on the metal surface, favoring at the same time the corrosion of refractories.
Moreover, it was found that the presence of Na in fluoride salts increases significantly the
aggressiveness of the aluminum alloy towards the refractory samples.

INTRODUCTION

One of the principal causes of degradation of refractory materials in the aluminum

industry is the chemical attack by molten aluminum (1). In the holding and melting
furnaces, this attack is particularly observed at the metal line, where the service
conditions are more severe (2). In fact, at the metal line, commonly called Bellyband
zone, the refractories are directly exposed to both molten aluminum and gaseous
atmosphere at high temperatures. At this triple contact, two major phenomena involving
molten aluminum may take place: 1 direct oxidation of the metal and 2 chemical
reduction of some refractory oxide compounds by molten aluminum. The latter may
occur either directly by molten aluminum below the metal line and/or through the oxide
layer above the metal line. The common feature of these chemical phenomena is
corundum formation. This corundum is a hard and adherent composite material,
containing molten metal channels. As it grows, it may extend inside the refractory lining
above the metal line. Therefore, it is almost impossible to remove this corundum deposit
without seriously spoiling the furnace lining.
To weaken the oxide layer during cleaning process, fluxes such as chlorides
and/or fluorides are added on the molten aluminum surface (3). Their main effect is to
promote metal/oxide separation by favoring the coalescence of metallic droplets within
the superficial dross. They may also act as a physical barrier against furnace atmosphere
by covering the metal surface and consequently preventing oxidation. However, the use
of such fluxing agents, especially those containing alkalis, may enhance the corrosion of
refractory materials in the Bellyband zone.
The purpose of this study is to analyze and to compare the effects of chlorides
versus fluorides on both oxidation and corrosion behavior of molten aluminum alloys.
Different aluminum alloys (Al-5wt.%Mg, Al-5.5wt.%Zn and the 7075 alloy), salts
(MgCl2, KCl, CaCl2, NaCl, MgF2, CaF2, NaF and cryolite) and aluminosilicate materials
are examined through a specific corrosion test, simulating the conditions at the metal line
in the aluminum treatment furnaces.

EXPERIMENTAL PROCEDURES

Materials and Sample Preparation

Four commercial aluminosilicate castables, having different silica and alumina
contents, are used in the present work. The chemical compositions of the tested products
are given in Table I.
The vibratory cast bricks made from the above materials are first dried, in an oven
at 110C for 24 hours, and then fired at 1200C over 5 hours in an electric furnace. After
firing, the bricks are cut to obtain cubic samples with a 7 cm-edge. A central cylindrical

cavity (diameter: 38 mm, depth: 50 mm) is then drilled in the cubes to obtain crucible
samples, which are required for CIREP laboratory Bellyband test. The cavity of samples
is filled, at room temperature, with about 100 g of aluminum alloys. Most of the tests
were performed with a laboratory prepared Al-5wt.%Mg alloy. Some tests were carried
out with Al-5.5wt.%Zn and commercial 7075 alloys, which contain 5.5wt.%Zn,
2.5wt.%Mg and 1.6wt.%Cu. The salts and small pieces of corundum (nuclei) are added
on the surface of the alloy, at room temperature. The addition of salts (normally less than
0.5 g) is made in the manner to cover approximately the of the surface of the metal and
thus to allow some direct contact between the metal and the atmosphere. As mentioned
before, chloride or fluoride-based salts (MgCl2, KCl, CaCl2, NaCl, MgF2, CaF2, NaF and
cryolite Na3AlF6) are used in this work.
Table I - Chemical Compositions (in wt.%) of the Tested Refractory Materials
Al2O3
Fe2O3
CaO
TiO2
Material
SiO2
C-8090
11.4
79.2
1.08
1.16
2.04
C-7080
14.4
78.3
0.99
4.07
1.65
C-6070
30.26
57.29
0.84
1.42
1.43
C-4060
48.0
42.6
1.26
5.38
1.54
* Presence of about 5wt.% of barite (BaSO4), as a non-wetting additive.

Others
5.12*
0.59
8.76*
1.22

Bellyband Corrosion Test

The CIREP Bellyband corrosion test (2) was designed to evaluate the effects of
salts on the metal oxidation and refractory corrosion of the samples (see set-up
configuration in Figure 1). In this test, the crucible samples are placed in a bottomloading-type furnace permitting to reach a temperature gradient from 1175C (testing
temperature) to about 900C (metal temperature) across the crucible sample (Figure 1).
According to the pre-established laboratory procedure, the surface of metal is disturbed
twice during the test. Once just after the test temperature is reached and a second time,
the day after. The test lasts 4 days and at the end of the test, the remaining metal in the
crucible is emptied to facilitate the longitudinal sample cutting. Finally, the cross section
of the samples is observed to evaluate the depth of penetration and/or corrosion of the
refractory along the cavity walls.
Thermogravimetric Analyses (TGA)
Using a CAHN TG171 thermo-balance, weight variations of metallic samples
were measured during a heat treatment. Samples were the cylinders of cast aluminum
alloys put inside an alumina crucible (diameter: 2 cm). Powders of salts were added to the
surface of the metallic cylinders, paying attention not to recover the whole surface. The
test lasts 48h at a constant temperature of 1175C. To reduce the loss of volatile alloying
elements, heating and cooling rates were as high as possible (15C/min.). With such a
rate, the testing temperature was reached in 90 minutes. Tests were performed only with
Al-5wt.%Mg containing the mentioned chloride and fluoride salts. Data of the TGA
measurements were continuously recorded during the tests. At the end of the test, the

samples were examined by means of optical and scanning electron microscopy. X-Ray
analyses are also performed to identify the nature of oxidation products.

Figure 1- Set-up for the CIREP Bellyband Corrosion Test

RESULTS AND DISCUSSION

Bellyband Corrosion Tests

The Bellyband corrosion tests, involving only the molten aluminum alloys
(without any salt) have been performed and were used as the reference for each refractory
material. Typical cross sections of the samples, which were significantly affected by
corrosion and/or oxidation during the Bellyband test, are illustrated in Figure 2. Based on
the visual examination performed on the cross sections of the samples, the degree of
refractory corrosion and metal oxidation were evaluated and reported in Table II.

C-7080
with Al-5wt.%Mg

C-4060
with Al-5.5wt.%Zn

C-6070
with 7075 alloy

Figure 2 - Cross Section of Some Samples, after the Bellyband Test

Table II: Evaluation of Corrosion and Oxidation of the Samples after

the Bellyband Tests
Test conditions
Alloy
Salt

Al-5wt.%
Mg

C-4060

C-6070

C-7080

C-8090

None
MgCl2
KCl
CaCl2
NaCl
MgF2
CaF2
NaF
Na3AlF6

Al-5.5wt.%
Zn

None
Na3AlF6

7075 alloy

None
Na3AlF6

: Refractory corrosion;

: Metal oxidation;

: Negligible;

: Moderate;

: Low;

: High;

: Extremely high

The above results underline the high tendency of Al-5wt.%Mg and 7075 alloys to
oxidation in the absence of fluxing agent. Moreover, among the tested alloys, Al5wt.%Mg appeared to be more aggressive against the refractory materials. Even the
refractories containing barium sulfate as a non-wetting agent (C-6070 and C-8090) were
affected by aluminum chemical attack. The fact that the non-wetting additives lose their
efficiency in protecting the refractories against Al-5wt.%Mg alloy attack after firing at
temperatures higher than 1000C, has already been discussed in the previous works (1,
4).
Based on its high tendency to oxidation and its aggressiveness against refractory
materials, the effects of fluxing agents were studied mainly with the Al-5wt.%Mg alloy.
Figure 3 and 4 show some cross sections of the samples, tested with this alloy containing
different chloride and fluoride salts, respectively. Referring to the data given in Table II,
it may be noted that the addition of salts reduced significantly the oxidation of the molten
aluminum alloys in most cases. This could be simply resulted from the reduction of the
contact surface of molten metal with air, due to the salt presence. Moreover, except the
cases involving NaF-based salts, the addition of the salts had no appreciable effect on the
corrosion of the refractory samples. However, when NaF or cryolite is used, severe
corrosion is observed in the aluminosilicate refractory samples. The aggressiveness of

these salts is likely related to the volatility of sodium and its high tendency to oxidation
and reaction with refractory oxide compound (5). To evaluate the reactivity of cryolite
with zinc as aluminum alloying element, the Bellyband tests with Al-5.5wt.%Zn and
7075 alloys were performed. Figure 5 compares the aspect of samples C-6070 with and
without Na3AlF5 in the case of zinc containing alloys. Similar to the Al-5wt.%Mg, the
addition of cryolite to aluminum-zinc alloys increased significantly the corrosion of the
four tested materials.

C-8090
With MgCl2

C-7080
With KCl

C-4060
With CaCl2

C-6070
With NaCl

Figure 3 - Cross Sections of Refractory Crucibles upon Test with

Al-5wt.%Mg and chlorides

C-7080
With MgF2

C-4060
With CaF2

C-8090
With NaF

C-6070
With Na3AlF6

Figure 4 - Cross Sections of Refractory Crucibles upon Test with

Al-5wt.%Mg and fluorides

Al-5.5wt%Zn
Without salt

Al-5.5wt%Zn
With Na3AlF6

7075 alloy
Without salt

7075 alloy
With Na3AlF6

Figure 5 - Cross Sections of the C-6070 Crucibles upon Test with Al-Zn alloys

For a given aluminum alloy, with or without salt, the oxidation aspect of molten
metal differs from a refractory crucible sample to another (see Table II). This suggests
that the chemical interaction between aluminum and aluminosilicate materials influences
the kinetics of metal oxidation. To better analyze the role of fluxing agents on the
oxidation behavior of aluminum alloys, thermogravimetric measurements were carried
out using alumina crucible, which is known as a highly resistant material against
aluminum chemical attack.
Thermogravimetric Analyses
To investigate the direct melt oxidation of Al-5wt.%Mg, with and without salts,
the weight variation of metallic samples as a function of heating process time was
continuously measured. Figure 6 shows the weight gain (dm) of an Al-Mg alloy,
expressed in mg/cm2, in the absence of salts. Based on the recorded thermogravimetric
measurements, the kinetics of metal oxidation was then calculated (in g/cm2.s). As
shown in Figure 7, the oxidation of Al-Mg alloys follows different characteristic steps
during heating process. According to the literatures first (6), an abrupt weigh gain,
represented by a narrow and sharp peak in Figure 7, is observed at around 850C. This
characteristic peak corresponds to a spinel (MgAl2O4) formation below a rapidly formed
thin external MgO layer, during heating ramp. Following the spinel formation, the
oxidation kinetics slows during a certain time, called incubation period. In the present
work, the incubation period extends from the end of the heating to the beginning of the
isothermal peak (at 1175C) and lasts about one hour (Figures 6 and 7). It is postulated
that, during this period, a slow thickening of the spinel layer, accompanied by the
development of an internal metallic network, takes place. Thus, the metal may infiltrate
through the spinel layer and reach the MgO layer. Once in contact with molten

aluminum, the magnesia layer becomes thermodynamically unstable and can be reduced
to form corundum (7). The incubation period ends with the nucleation of Al2O3 nodules,
between the spinel and magnesia layers. Several studies have shown that the incubation
time could be reduced, for instance by doping the metal surface with specific chemical
species such as silica powders (6-8). In the present investigation, a reduction of the
incubation period was observed through a complementary test using a little piece of
corundum as surface dopant. Following the incubation period, a second rapid weight
gain, characterized by the large peak in the kinetics curve of Figure 6, occurs. According
to the literatures (6, 7), this second large and extended peak corresponds to the growth of
an Al2O3/Al composite material, sandwiched between MgO and MgAl2O4 layers. It
seems that this ceramic/metal composite continues to grow, in a quasi-steady state, as
long as the molten alloy is somewhere in contact with bulk spinel; i.e., a continuous metal
supply is assumed. According to the theoretical analyses (9), the corundum growth rate in
this quasi-steady state is governed by the oxygen partial pressure and temperature, as
expressed in the following equation:
VOxidation = 3.6108 PO21/4exp-370(kJ/mol)/RT
In the present experimental conditions (PO2=0.2atm., T=1448K) , the calculated
value for the theoretical oxidation rate is 11 g/cm2.s. This value, reported in the
experimental curves, is used for comparison. The average experimental value established
from the slope of the oxidation curve in Figure 5 is 20 g/cm2.s.; almost two times higher
than the theoretical value.

Thermogravimetric analyses - Weight variation

Al-5%wt Mg without any salt addition - Without air flow
850

1400

800

750
1200

700
650

Temperature T (C)

550
500

dm /dt

800

450
400
350

600

300
250
400

Weight variation dm (mg/cm2)

600

1000

200
150
100

200

50
0
0

-50
0

10

15

20

25

30

35

40

45

50

Time t (h)

Figure 6 - Weight Gain of an Al-5wt.%wtMg without any Salt Addition

Thermogravimetric analyses - Oxidation kinetics

Al-5%wt Mg without any salt addition - Without air flow
60

1400

55

T
1200

50

Temperature T (C)

40
35

Bulk corundum growth

dm/dt

800

30
25

600
20

Theoretical value = 11g/cm2.s

400

15

Weight variation kinetics (g/cm2.s)

45

Spinel formation peak

1000

10
5

200

0
0

-5
0

10

15

20

25

30

35

40

45

50

Time t (h)

Figure 7 - Oxidation Kinetics of the Al-5wt.%wt Mg Derived from Figure 5

The average oxidation rate of the Al-5wt.%Mg samples, containing different salts,
is reported in Table III. An important slowdown, about one order of magnitude, in the
oxidation kinetics of the alloy is observed especially with the addition of chlorides.
Moreover, the oxide colors of thermo-balance samples with chloride salts appear visibly
whiter compared to the others, as shown in Figure 8. In the case of MgF2 salt addition, an
extensive MgO whiskers formation above the thermo-balance alumina crucible was
observed (details in the next section). It should be noted that with this salt, the reduction
in the oxidation rate of samples was less significant, compared to the other salts.
Table III: Oxidation aspects of Al-5wt.%Mg samples during thermo-balance tests
Salt
Oxidation
rate*
(g/cm2.s)
Oxide color
Observation

None

MgCl2

KCl

CaCl2

NaCl

MgF2

CaF2

NaF

Na3AlF6

20

11

Dark gray

White
& gray

White

White

White

Dark gray

Dark gray

Dark
gray

Dark gray

NaCl was
decomposed
during
heating

Extensive
MgO
whiskers
formation

Low MgO
whiskers
formation

Metal
almost
totally
oxidized

* Average rate established in the first fifteen hours of the thermo-balance tests.

Saltfree

MgCl2

NaCl

MgF2

NaF

Na3AlF6

Figure 8 - Appearance of Al-5wt.% Mg Samples after the Thermo-balance Tests

Figures 9 and 10 illustrate two typical examples of the oxidation kinetics of

samples with chlorides and fluorides, respectively. In both cases, the narrow peak
corresponding to spinel formation during heating was observed. However, with NaCl, the
weight gain due to spinel formation is partly reduced by salt decomposition before
reaching the testing temperature. The major difference between chlorides and fluorides is
the bulk corundum growth at the isothermal test temperature (1175C). With chlorides,
the kinetics of corundum growth is continuously reduced up to reach a plateau (steadystate) of about 0.3 g/cm2.s, whereas with fluorides, the steady-state of aluminum
oxidation (also nearly 0.3 g/cm2.s) is frequently disrupted by irregular sequences of
progressive increase and then abrupt decrease in weight (Figure 10). These oscillations in
the oxidation behavior may be interpreted by the sudden flow of variable amounts of
molten metal to the oxide surface and their subsequent oxidation. This should also
explain the relative higher values of oxidation rate (Table III) obtained with fluorides in
comparison with chlorides.

Thermogravimetric analysis - Oxidation kinetics

Al-5%wt Mg + NaCl - Without air flow
20

1400

1200

15

Theoretical value = 11g/cm .s

5
1000

Temperature T (C)

dm/dt

-5

Steady-state

800

-10
-15
600
-20
-25

400

-30

Weight variation kinetics dm/dt (g/cm2.s)

10

-35

200

NaCl decomposition

-40

-45
0

10

15

20

25

30

35

40

45

50

Time t (s)

Figure 9 - Oxidation Kinetics of the Al-5%wt Mg Sample with NaCl

Thermogravimetric analysis - Oxidation kinetics

Al-5%wt Mg + NaF - Without air flow
1400

55

T
1200

1000

Temperature T (C)

35

dm/dt

800

25
600

Theoretical value = 11g/cm2.s

15

400

Weight variation kinetics dm/dt (g/cm2.s)

45

5
200

-5
0

10

15

20

25

30

35

40

45

50

Time t (h)

Figure 10 - Oxidation Kinetics of the Al-5%wt Mg with NaF

SEM Analyses
To better understand the difference between the effects of chlorides and fluorides
on the oxidation behavior of Al-Mg alloys, some thermo-balance samples were observed
and analyzed by a scanning electron microscope. Figure 11 illustrates the appearance of

the external oxide surfaces, i.e. in direct contact with atmosphere, of the samples
containing NaCl and CaF2. The X-Ray element analyses performed on these sample
surfaces are given in Figure 12. The sample containing NaCl seems to be covered by a
continuous MgO layer, having visibly a smooth appearance, whereas with CaF2, the
oxide surface appears more tortuous and contains a large amount of aluminum. This may
confirm the earlier suggestion concerning the periodic flow of molten aluminum from
bulk to the oxide surface and its direct exposure to the atmosphere during thermo-balance
tests of fluoride-doped samples. In other words, with fluorides, the external MgO layer is
less protective in comparison with chlorides. This may also explain why with fluorides
the corrosion of aluminosilicate samples appeared more pronounced than with chlorides.

NaCl

CaF2

Figure 11 - Surface of Salt Surface-doped Al-5wt.%wtMg Samples

NaCl

CaF2

Trace of residual salt

Figure 12 - X-Ray Spectra of the Oxidized Sample Surfaces in Presence of NaCl and
CaF2

Figures 13 and 14 show the cross sectional microstructures of the samples doped
with CaCl2 and cryolite, respectively. The X-ray mapping of the three principal elements,
i.e. Al, Mg and O, constituting the oxide layers of the samples are also shown in these
figures. The overall outcomes of the oxide layers of these salt-doped samples are in good
agreement with the general observations and analyses of aluminum-magnesium alloy
oxidation. The presence of an intermediate spinel layer, in contact with metal bath, and
bulk corundum is quite visible in both cases. The thickness of the oxide layers with
Na3AlF6 (about 2mm) is 4 times larger than that with CaCl2 (about 500m). This is also
in good agreement with the above thermogravimetric analyses, showing an average
weight gain rate of about 4 times higher with fluorides in comparison with chlorides
(Table III). The main difference between the two microstructures is the presence of a
thick and relative dense MgO external layer, in the case of CaCl2 addition. The thickness
of this magnesia layer (50 to 100m) is at least one order of magnitude higher than that
(1 to 10m) usually observed during high temperature oxidation of aluminummagnesium alloys (7-9). The formation of such a thick magnesia layer may also explain
why with chloride surface doping, the kinetics of oxidation was more sluggish. This may
also suggest that the effects of salt addition are not limited only to a physical barrier role,
but it may also influence the microstructure of the oxide layers.
Finally, Figure 15 shows the micrographs of whiskers and droplets formed above
the sample with MgF2. Using X-ray analyses, these whiskers were identified as MgObased crystals, which apparently have been nucleated and grown from magnesium vapor
phase during thermo-balance test.

Magnesia

Corundum

Spinel
Aluminum
Al

Mg

Figure 13 - Microstructure and X-Ray Mapping of the Oxide Products Formed on the
Surface of an Al-5wt.%Mg Sample Doped with CaCl2.

Corundum

Spinel
Aluminum
Al

Mg

Figure 14 - Microstructure and X-Ray Mapping of the Oxide Products Formed on the
Surface of an Al-5wt.%Mg Sample Doped with Cryolite

Droplets

Whiskers

Figure 15 - Microstructure of MgO Whiskers Formed Above an Al-5wt.%Mg Sample

Doped with MgF2.

CONCLUSION

The addition of chlorides and fluorides surface dopants affects differently the
oxidation and refractory corrosion generated by aluminum alloys. From the corrosion
point of view, the use of chlorides has no significant effects, while with NaF-based salts
the metal attack could be critically accelerated. The aggressiveness of sodium fluoride
versus refractory materials were observed in both aluminum-magnesium and aluminumzinc alloys. The chlorides also appeared to be more efficient than fluorides in the

protection of the aluminum-magnesium alloys from oxidation. It seems that use of

chlorides favors the formation of a thick protective MgO layer.

ACKNOWLEDGEMENTS
The authors are grateful to the Natural Sciences and Engineering Council of Canada for
its financial support within the research activities of the NSERC-REFRAL Chair. The
authors also acknowledge the industrial partners of this Chair, i.e., Alcan International,
Pechiney CRV, RHI Canada and Lafarge Refractories, for their financial and technical
support.

REFERENCES

1. S. Afshar and C. Allaire, The Corrosion of Refractories by Molten Aluminum,

JOM, Vol. 48, May 1996, 23-27.
2. S. Quesnel, S. Afshar and C. Allaire, Corrosion of Refractories at the Bellyband of
Aluminum Melting and Holding Furnaces, Light Metals, 1996, 661-667.
3. A. Silny and T.A. Utigard, Interfacial Tension between Aluminum, Aluminum based
Alloys and Chloride-Fluoride Melts, Light Metals, 1997, 871-878.
4. S. Afshar and C. Allaire, Furnaces: Improving Low Cement Castable by NonWetting Additives, JOM, Vol. 53, August 2001, 24-27.
5. C. Allaire and P. Desclaux, Effect of Alkalis and of a Reducing Atmosphere on the
Corrosion of Refractories by Molten Aluminum, J. Am. Ceram. Soc., Vol. 74, No.
11, 1991, 2781-2785.
6. O. Salas, V. Jayaram, K.C. Vlach, C.G. Levi, and R. Mehrabian, Early Stage of
Composite Formation by Oxidation of Liquid Aluminum Alloy, J. Am. Ceram. Soc.,
Vol. 78, No. 3, 1995, 609-622.
7. S. Antolin and A.S. Nagelberg, Formation of Al2O3/Metal Composite by the
Directed Oxidation of Molten Aluminum-Magnesium-Silicon Alloys: Part I,
Microstructural Development, J. Am. Ceram. Soc., Vol.75, No. 2, 1992, 447-454.
8. H.S. Park and D.K. Kim, Effect of Silica Surface Dopant on the Formation of
Alumuna/Aluminum Composite by the Directed Metal Oxidation of an Aluminum
Alloy, J. Am. Ceram. Soc., Vol. 84, No. 11, 2001, 2526-2530.
9. A.S. Nagelberg, S. Antolin, and A.W. Urquhart, Formation of Al2O3/Metal
Composite by the Directed Oxidation of Molten Aluminum-Magnesium-Silicon
Alloys: Part II, Growth Kinetics, J. Am. Ceram. Soc., Vol. 75 No. 2, 1992, 455-462.