55

A MODIFICATION OF PRAGER'S HARDENING RULE*

BT
HANS ZIEGLER
Eidgenossische Technische Hochschule, Zurich

1. Introduction. The response of a rigid-work-hardening material can be described

by
a) an initial yield condition, specifying the states of stress for which plastic flow
first sets in,
b) a flow rule, connecting the plastic strain increment with the stress and the stress
increment,
c) a hardening rule, specifying the modification of the yield condition in the course
of plastic flow.
It is customary to represent the yield condition as a surface in stress space, convex
[1] and initially containing the origin. The current yield conditions for a metal are
those of v. Mises [2] and of Tresca [3]. The flow rule generally accepted [4, 1] is also
due to v. Mises [5]. It states that the strain increment vector lies in the exterior normal
of the yield surface at the stress point. As to the hardening rule, there are mainly two
versions in use. The rule of isotropic work-hardening given by Hill and Hodge [6, 7]
assumes that the yield surface expands during plastic flow, retaining its shape and
situation with respect to the origin. Another rule, developed by Prager [8], assumes
that the yield surface is rigid but undergoes a translation in the direction of the strain
increment.
The rule of isotropic work-hardening does not account for the Bauschinger effect
observed in the materials in question. Prager's hardening rule accounts for this effect.
However, as Perrone and Hodge Jr. [9] have shown in special cases and Shield and the
author [10] in a general investigation, Prager's hardening rule is not invariant with
respect to reductions in dimensions possible in almost any applications. In other words:
if the yield surface in 9-space uik moves in the direction of the exterior normal at the
stress point, the two-dimensional yield locus, e.g., in plane stress vx , <rydoes not do so.
In certain cases, e.g., if only <rxand txv are different from zero, the Tresca yield locus
in the plane <jx, txv even deforms.
It is clear that the physical consistency of Prager's rule is not affected by the phe-
nomena last mentioned. However, they complicate the application of the rule particularly
in cases which otherwise would be simple enough to lend themselves to a complete
treatment. On the other hand, the investigations of Shield and the author have shown
that, under v. Mises' yield condition at least, the yield surface, in all special cases,
according to Prager's rule moves in the direction of the radius connecting its center
with the stress point. This suggests that a corresponding modification of Prager's rule
might simplify certain problems. In the following sections such a modification is formu-
lated, investigated and compared with Prager's rule.
2. The modified hardening rule. Let us consider an element of a rigid-work-harden-
ing solid, referred to an orthogonal coordinate system xt . The state of stress of this

*Received Jan. 2, 1958.

56 HANS ZIEGLER [Vol. XVII, No. 1

element can be represented by a stress point P in a 9-space aik with origin 0. In this
space, the initial yield surface is represented by an equation
F(aik) = k2 = const. (2.1)
In the following, for simplicity, attention will be confined to initially isotropic materials
for which the form of the function F is invariant with respect to a rotation of the stress
state. An initially anisotropic material can be treated in an analogous manner.
The hardening rule suggested by Prager assumes that during plastic deformation
the yield surface moves in a translation. After a certain amount of plastic flow, it is
given by
F(<rik — aik) = k2, (2.2)
where the tensor aik represents the total translation. Because aik is not necessarily the
isotropic tensor 8ik , where 5ik is the Kronecker delta, the material becomes anisotropic as
a result oj the hardening process. Accordingly, direction is important, and we shall fix
the coordinate system x{ with respect to the element, small deformations being assumed.
In the space <rik, the aik represent the radius vector of the point C, which initially
was at the origin and which in the following will be referred to as the center of the yield
surface. Due to the flow rule of v. Mises, the plastic strain increment deik , considered
as a vector in the same space, lies in the exterior normal of the surface (2.2) at P. Thus,
it is represented by

dett = d\, d\ > 0. (2.3)

OCTik

The definition of a Prager-hardening material is completed by assuming that the surface

(2.2) moves in the direction of deik ; more explicitly
daik = c deik , (2.4)
where c is a constant characterizing the material.
Instead of (2.4), let us assume
daik — (<rik — aik) dfi, d\x > 0, (2.5)
i.e. that the yield surface still moves in a translation, but in the direction of the vector
CP connecting the center of the yield surface with the stress point (Fig. 1). This rule

0
Fig. 1. Hardening rule and flow rule for a linear work-hardening solid.

is a modification of Prager's law, physically acceptable since both sides of (2.5) are
tensors of the second order.
The scalar dfi in (2.5) is determined by the condition that P remains on the yield
surface in plastic flow. If in 9-space the summation convention is adopted, this condition is
1959] A MODIFICATION OP PRAGER'S HARDENING RULE 57

dF
{d<rik— daik) -— = 0, (2.6)
0(T ik

and from (2.5) follows at once

, _ ('dF/do-jj) dan fo 7n
" (ff*i - <*>,)dF/d<rkl' U'7j

Since (2.4) has been replaced by (2.5), the vector OC no longer represents the total
strain. This is a serious drawback of our modification of Prager's rule. On the other
hand, we gain a substantial advantage: d\ in (2.3), i.e. the magnitude of the strain
increment remains free and can still be established as a suitable function of the stress,
the stress increment and the stress history. In other words: it is possible to adjust the
modified hardening rule (2.5) to any kind of hardening law in simple tension and compression.
The simplest way to dispose of d\, i.e. to complete the flow rule, is to assume that
the vector cdtik is the projection of d<rik (and thus of daik) on the exterior normal of the
yield surface. This corresponds to the procedure familiar from Prager's rule, and it will
turn out later on that on this basis the results of either rule coincide in many cases. If
d\ is fixed in this way, the total strain is represented by the sum of the infinitesimal
translations of the yield surface in the direction of its exterior normal at the stress point.
Since
dF
(d(xik - c deik) -— = 0, (2.8)
00" ik

we obtain from (2.3)

c (dF/dakl) (dF/dakt) ^

By means of the hardening rule (2.5), (2.7) and the flow rule (2.3), (2.9) the material
is completely defined. It is easy to see that these rules are a generalization to complex
states of stress of a linear hardening law in tension and compression, Fig. 2, which
exhibits a Bauschinger effect. Moreover, by assuming that c is not a constant, but a

G0+C,E

J-w
e0-c,c

Fie. 2. Linear work-hardening in simple tension and compression.

suitable function of the stress history, e.g. of the distance OC or of the dissipation work,
it is possible to adapt our rules to any material with a given non-linear hardening law
in simple tension and compression.
For most of the following considerations it will not be necessary to restrict ourselves
to linear work-hardening materials, i.e. to constant values of c.
58 HANS ZIEGLER [[Link], No. 1

3. Some general properties. In an initially isotropic solid the yield function takes
the form
F(aik) = <?[/i((ra), I2(cra), I3(<Tik)], (3.1)
where
I\ = Cii , I2 = i<TijTji , 13 = jk(Tki (3.2)

are the invariants of the stress tensor. If the initial yield is independent of the mean
normal stress,
+ /38ik) = F(aik), (3.3)
where 0 is an arbitrary scalar. When plastic flow has set in, the yield function becomes,
on account of (2.2) and (3.1),
F(ffik — ««) = G[Ix{<Ja — Ola), h(<rik ~ «a), /3(o"a — a,-*)]. (3.4)
From (3.3) follows that the values of (3.4) remain unchanged when <rikis replaced by
<xik+ /35i4 : if yield is initially independent of the mean normal stress, it remains so.
On account of (3.4), the flow rule (2.3) becomes

ri — (— d^» 1 d/2 , dG §h_\ A (n

dtik la/, dait + dh daik+ dh daJ dX' (3'5)
Carrying out the differentiation of the invariants, we obtain

j r . dG, » . dG, ., . "1, . .

= LaT" a/2 — ~dh ~ <Xl'^<r'k
~ a'k' J ' ^
Let us assume now that the physical coordinate system initially coincides with the
principal axes of stress. Then we have first
crit = 0, (i k) and au = 0. (3.7)
From (2.5) and from (3.6) follows
daik = 0, (i 7^ k) and deik = 0, (i 5* k). (3.8)
Since the material is isotropic at the outset, this means that the strain increment tensor
deik and the tensor daik are coaxial with the stress tensor aik . The relations (3.8) remain
valid if the second assumption (3.7) is replaced by the weaker assumption
"it = 0, (i k). (3.9)
It follows that, if the -principal axes of stress remain fixed in the element from the start,
the strain increment tensor and thus the strain tensor remain coaxial with the stress tensor.
If the principal axes of stress rotate, (3.8) holds only in a first step, provided the
principal system of stress is used as the physical coordinate system. If (3.8) shall hold
in a second step, the coordinate system must be rotated between the first step and the
second one. This rotation, however, violates (3.9): due to the anisotropy caused by strain
hardening, the strain increment tensor is in general not coaxial with the stress tensor.
Many problems of practical importance can be treated in a stress space of less than
9 dimensions. In certain cases, e.g., a 3-space defined by the principal stresses is useful.
From our last result follows, however, that this 3-space is inadequate where the principal
axes of stress are not fixed in the element.
1959] A MODIFICATION OF PRAGER'S HARDENING RULE 59

The properties obtained in this section apply without exception also for a material
obeying Prager's rule [10].
4. Treatment in subspaces. On account of the symmetry of the stress and strain
tensors, the problem may as well be treated in 6-space. It is convenient here and par-
ticularly for the subsequent specializations to denote the physical coordinates by x, y, z,
the stresses by crx, • • • , r,,, • • • , and the strains by ex , where the dots
indicate cyclic permutations.
In the new notations the yield condition (2.2) reads

F(<rx — ax , ■ • • , tvz — avz , • • • , tzv — azu ,•••) = k2, (4.1)

where tvz , rzv , • • • have to be considered as independent variables. The flow rule (2.3)
becomes

dex = d\, ■■■ , divz = d\, ■■■ , dezv = d\, ■■■ , (4.2)
dcx o Tyz drzy

and the hardening rule (2.5) takes the form

docx ((Tj; o^) dfij ' j davz (.Tyz d/j.) '
(4.3)
dazv = (r2„ — azv) dp, ■ • • .}

Treatment in 6-space, however, requires the elimination of the stress components

rty (= ryz), • ■• , of the strain components ezy (= e„2), • • • , and of the displacements
azv, ■■■ , which, on account of (4.3), are equal to the displacements ayz, • • • . On account
of the symmetry of the stress tensor

> *** > , • • • , tzv , • • •) = f(ax , • • • , r„ , • • •)• (4.4)

Thus, the yield surface in 6-space is given by

/(°"z ax J ' ] 7"1/2 avz J ' ' ') (4 5)

F\.GX OtX f ' ' ' , Tyt OtyZ , ) Tzy OLgy , *] A/ .

From (4.2) and (4.5) we obtain

dex — d\, ••• , dyvz = 2 dtuz = d\, ••• . (4.6)

a <rx dTyz

This is the well known result that the flow rule of v. Mises remains valid in 6-space, if
the state of strain is represented by the engineering components ex , • • ■ , yyz , • • • .
Further, from (4.3) follows
dax = (crx — ax) dn, ■■ • , day. = (tvz — a„) dp, . (4.7)

Thus, the hardening rule (2.5) applies without modification also in 6-space.
In many practically important cases, some of the stress components are identically
zero. Starting once more in 9-space, let us denote the stress components present by
c'k , the zero ones by a"k . The initial yield condition is then
F{afik, all = 0) = = k\ (4.8)
If we are not interested in the strains e"k corresponding to the zero stresses <s\'k, we may
treat the problem in a subspace u'ik . Here defines a new yield surface.
60 HANS ZIEGLER [Vol. XVII, No. 1

On account of the hardening rule (2.5)

aii = 0. (4.9)
Thus, the yield surface, after plastic flow has set in, is given by
F{c[k - a'tt , - <x[l = 0) = H(a'th - a'ik) = k\ (4.10)
It follows that also in the subspace <r'ikthe yield surface moves in a translation.
On account of (4.10)

deU--frd<Tik dk - d(Tik
dk. (4.11)
Thus, the flow rule remains valid in any subspace. Of course, it supplies only the strain
components e'ikdefined in this subspace, although the e'k, too, may be different from zero.
Finally, from (2.5) follows
da'a = Wik ~ a'k) d/i, (4.12)
i.e., also the hardening rule (2.5) remains valid in any subspace. This is another advantage
in comparison with Prager's rule which in most subspaces applies only in a modified
form [10].
In the next sections, we shall discuss a few special states of stress particularly import-
ant in applications. We shall restrict ourselves to materials where yield is independent of
the mean normal stress. Here, as in (3.3),

/((7*x ax + ' , Tve ayz , ' •) /(o"z OLx f ''' j Tuz <xyt , •). (4.13)

5. Plane strain. Here,

Tut Tzx 0, 0, (5.1)

by definition. From (4.9) follows av, = a!X = 0. Thus, the yield function has the form

g{<rx — Ctx , <rv — a„ , <r, — a, , txv — axv). (5.2)

On account of (5.1) and the flow rule (4.6), <jzmust be absent from (5.2). Using (4.13),
we finally obtain the yield condition

h(<rx — a'x , cy — a'y , rxu — axv) = fc2, (5.3)

where
ocx = ax — a, and a'y = av — a, .

If the material obeys v. Mises' yield condition, we have initially

{<rx~ <02 + 4r2„ = | Co , (5.4)

where <r0is the yield limit in simple tension or compression. The yield surface therefore
is the elliptic cylinder

[(<r*— aQ — yiTy— a'y)]2 + 4(tx„ — aty)2 = | <rl (5.5)

with its axis parallel to the line bisecting the angle of the first quadrant and with semi-
axes (2/3)1/2o-0) (l/3)1/2<r0 .
1959] A MODIFICATION OF PRAGER'S HARDENING RULE 61

If Tresca's yield condition holds, we have, instead of (5.4),

(.<rx — ov)2 + 4= a20 . (5.6)

The yield surface is again an elliptic cylinder

[(<rx — a'x) — (<r„ — a'v)]2 + 4 (r„ — aXy)2 = al (5.7)

with its axis in the same direction, but with semiaxes <ra/21/2, <r0/2.
Subcase a: If txv = 0, (4.9) yields axv = 0. The problem can be treated in a plane
<tx, <tv, and the yield locus is the strip (Fig. 3) obtained by bisecting the cylinder (5.5)
or (5.7) parallel to the plane <rx, (rv.

([Link])
^ /1
y\ YSeo (Tresca)

Fig. 3. Yield locus in plane strain with txu = 0.

Subcase b: If <rv= 0, (4.13) can be used once more, and the problem can be treated
in a plane <rx, txv . The yield locus is the ellipse (Fig. 4) obtained by bisecting the cylinder
(5.5) or (5.7) parallel to the plane <rx, txv . Its semiaxes are 2<r0/31/2, o-0/31/2for v. Mises'
yield condition and tr0 , o-0/2 for the condition of Tresca.

Fig. 4. Yield loeus in plane strain with <ry= 0.

Since the length of the cylinder (5.5) or (5.7) is infinite, it does not matter whether
its translation is given by the vector da in the direction CP or by its projection on the
exterior normal. It follows that for a linear work-hardening law [c in (2.9) constant]
the results obtained here coincide with those supplied by Prager's hardening rule.
6. Plane stress. Here, by definition,
<r. = tv. = r„ = 0. (6.1)
From (4.9) follows a, = av, = atx = 0. Thus, the yield surface has the form

g(irx — ax , <TV— av , txv — axu) = k2. (6.2)

62 HANS ZIEGLER [[Link], No. 1

If the material obeys v. Mises' yield condition, we have initially

<rl + <rl — <tx<7v
+ 3t2„ = <to (6.3)
and thus
(c* — a*)2 + (,<r» — aty)2 ~ (<rx — ax)(trv — a,) + 3(rIV — axv)2 = <r2 . (6.4)

The yield surface is an ellipsoid with semiaxis 21/2<r0in the direction of the line bisecting
the first quadrant in the plane ax , av , with semiaxis (2/3)1/2<r0 in the direction of the
line bisecting the second quadrant and o-0/31/2 in the direction of the axis txv .
If Tresca's condition holds, we have initially the yield limits
[(<7* — <r„)2 + 4r2„]1/2 = <T0 (6.5)

and
§ | <sx + CT„± [(<rx — ct„)2 + 4r2„]1/2 | = c0 . (6.6)

Thus, the yield surface is an elliptic cylinder with the same orientation as v. Mises
ellipsoid, closed by two elliptic cones.
Subcase a: If tx„ = 0, (4.9) supplies ax„ = 0, and the problem can be treated in a
plane <rx, cr„ . If v. Mises' yield condition (6.4) applies, the yield locus is the ellipse of
Fig. 5 with the equation
(<rx — ax)2 + (o-„ — <*„)2 — {<TX— ax)(ay — a„)fi= al . (6.7)

Fig. 5. v. Mises yield locus in plane stress with txv = 0.

If the material obeys Tresca's yield condition (6.5), (6.6), the yield locus is the hexagon
of Fig. 6.

p i/c/a
/
/
/
¥

Fig. 6. Tresca yield locus in plane stress with Txy = 0.

1959] A MODIFICATION OF PRAGER'S HARDENING RULE 63

Subcase b: If ay = 0, (4.9) yields <xy= 0, and the problem can be treated in a plane
<rx, txj ■ The yield locus is the ellipse of Fig. 4 with semiaxes <r0, o-0/31/2 in v. Mises'
case and <r0, o-0/2 in the case of Tresca.
If P lies in a corner or vertex of the yield surface, the strain increment exhibits the
same indeterminacy in direction as in a perfectly plastic material. In the lower right
corner of the Tresca hexagon of Fig. 6, e.g., the vector dt lies anywhere in the shaded
region defined by the normals of the adjacent sides. However, P remains only in the
corner or vertex if the yield surface moves with P, i.e. if the stress increment d<i has the
direction CP.
7. Another special case. In certain cases, e.g. in a cylinder subjected to torsion and
and tension, we have

<rx = ov = rxy = 0. (7.1)

From (4.9) follows ax — a„ = axy = 0. Thus, the yield function has the form

(]{(Tz OLz j Tyz &yz ] Tzx &zx) ^ • (y *2)

If the material obeys v. Mises' yield condition, we have initially

crl + 3 (rl, + T«) = (To (7.3)
and thus
(<yz — a*)2 + 3(r„j — ayz)2 + 3(r„ — azx)2 = al . (7.4)

The yield surface is an ellipsoid of revolution with semiaxes <t„, <r0/31/3.

If Tresca's yield condition applies, the principal stresses are initially
0-1 = o, 0-2,3 = il<Tz ± {al + 4(t2z + tL)}1/2]. (7.5)

The maximum shear stress is (<r2— c3)/2; hence, we have, instead of (7.4),
(tr2 — oiz)2 + 4(r„2 — ayz)2 + 4(r„ — a„)2 = , (7.6)

i.e. an ellipsoid of revolution with semiaxes <t0, <r0/2.

Subcase a: If az = 0, (4.9) supplies az = 0. The problem can be treated in a plane
Tyz> tzx . The yield locus is a circle of radius <x0/3l/2in v. Mises' case and c0/2 in the case
of Tresca.
Subcase b: If ryz = 0, (4.9) yields ayz = 0, and the problem can be treated in a plane
<rz, tzx . The yield locus is an ellipse as in Fig. 4 with semiaxes <r0, <r0/31/2 in v. Mises'
case and <r0,a0/2 in the case of Tresca. It is clear that this result, apart from the difference
in notation, is the one of Subcase b in Sec. 6.
8. Discussion. By suitable linear transformations of the stresses (and of the aik)
the yield surfaces and yield loci obtained in Sees. 5 through 7 could be simplified. Ellip-
soids can be transformed into spheres, ellipses into circles and so on. The hardening
rule (2.5) is not affected by such a process. However, if the flow rule (2.3) is to apply in
the new variables, it is necessary to transform the strains [10, 11] simultaneously.
At the present stage of research, it seems hopeless to expect a decision by experiment
between the hardening rules confronted here. Comparisons must be based, therefore, on
purely theoretical reasoning.
It has been pointed out in Sec. 5 that in plane strain the results obtained for a linear
work-hardening material by the hardening rule (2.5) are the same as those supplied by
64 HANS ZIEGLER [[Link], No. 1

Prager's rule. A comparison with [10] shows that the situation is the same in plane
stress and in the stress state of Sec. 7, provided that the material obeys v. Mises' yield
condition. So far it does not matter which hardening rule is used. If, however, a material
obeying Tresca's yield condition is subjected to plane stress or to the state of stress of
Sec. 7, the response depends on the hardening rule. Under Prager's rule, the yield surface
deforms in most cases. Under the rule (2.5), such deformations do not occur, since,
according to a statement in Sec. 4, the hardening rule (2.5) applies without modification
in any subspace. This is a definite advantage of the rule (2.5).
Another point has been raised in-Sec. 2: the possibility of adjusting the hardening
rule (2.5) to arbitrary non-linear hardening laws in simple tension and compression.
This advantage, however, is obtained at the expense of the geometrical interpretation
of the total strain. If one is prepared to renounce this possibility, one clearly is in a
position also to adjust Prager's rule to non-linear laws. This point, therefore, is hardly
of importance.
A serious drawback of the hardening rule (2.5) lies in the indeterminacy of the strain
increment in a corner or vertex of the yield surface. This problem arises already if the
Tresca yield condition is applied to a perfectly plastic solid.
Let us assume that a perfectly plastic cylinder with axis x is subjected to simple
tension. In plastic flow, the stress point has the position P in Fig. 7. The exterior normal

Fig. 7. v. Mises and Tresca yield in simple tension.

at P of v. Mises' ellipse has two components with the ratio 2: (—1). Hence, the contrac-
tion of the cylinder in the direction y is half the elongation in the direction x and thus
the same as the contraction in the direction z, since the volume change is zero. In other
words: if v. Mises' yield condition applies, the deformation of the cylinder exhibits the
same rotational symmetry with respect to the axis x as the state of stress. However,
if the material obeys Tresca's yield condition, the direction of dt is free within the shaded
angle formed by the normals in P of the sides 1 and 2 of the Tresca hexagon. In view
of the incompressibility of the material this means that under Tresca's yield condition
the deformation may be arbitrarily unsymmetric. There is no reason why the cylinder
should not contract, e.g., in the direction y alone. Experimental evidence seems to decide
against Tresca's yield condition and its associated flow rule. It is clear, however, that
in experiments an exactly linear state of stress cannot be realized. In any event, if we
accept the Tresca yield condition together with the flow rule of v. Mises, we have to
admit the possibility of an arbitrarily asymmetric deformation in simple tension and
compression.
1959] A MODIFICATION OF PRAGER'S HARDENING RULE 65

The difficulty considered here is not present in a material work-hardening in accord-

ance with Prager's rule. Here, the strain increment in a corner of the Tresca hexagon
is uniquely determined [10] by the stress increment, and the ratio dex/dev is 2: (—1) in
simple tension, no matter which yield condition is used. However, a material hardening
according to the rule (2.5) exhibits the same indeterminacy in dt as a perfectly plastic
material. In this respect the rule (2.5) is not quite satisfactory.
We intend to compare the two rules in various examples. So far, it seems that the
rule (2.5) is inferior to Prager's hardening rule from a physical point of view, but easier
to handle in certain classes of applications.

References
1. H. Ziegler, An attempt to generalize Onsager's principle, and its significance for rheological problems,
ZAMP 9b, 748 (1958)
2. R. v. Mises, Mechanik der feslen Korper im plastisch deformablen Zustand, Goettinger Nachrichten,
Math. phys. Klasse 1913, 582 (1913)
3. H. Tresca, Mimoire sur I'ecoulement des corps solides, Mem. pres. par div. savants 18, 733 (1868)
4. D. C. Drucker, Some implications of work hardening and ideal plasticity, Quart. Appl. Math. 7,
411 (1950)
5. R. v. Mises, Mechanik der plastischen Formanderung von Kristallen, Z. angew. Math. Meeh. 8,
161(1928)
6. R. Hill, The mathematical theory of plasticity, Oxford, 1950
7. P. G. Hodge, Jr. The theory of piecewise linear isotropic plasticity, IUTAM Colloquium Madrid
1955, "Deformation and Flow of Solids," Berlin, 1956
8. W. Prager, The theory of plasticity: a survey of recent achievements, Proc. Inst. Mech. Engrs. 169,
41 (1955)
9. N. Perrone and P. G. Hodge Jr., Applications of a consistent theory for strain hardening plastic solids,
PIBAL Report 403, Sept. 1957, Polytechnic Institute of Brooklyn, Brooklyn, N. Y.
10. R. T. Shield and H. Ziegler, On Prager's hardening rule, ZAMP 9a, (1958)
11. W. Prager, Probleme der PlastizitMstheorie, Basel, 1955