ONE-DIMENSIONAL THERMOMECHANICAL
PHASE TRANSITIONS WITH NON-CONVEX
POTENTIALS OF GINZBURG-LANDAU TYPE
By
Jirgen Sprekels
IMA Preprint Series # 505
April 1989‘This paper will appear in the IMA
‘Volume Proceedings for the March 1989
workshop on Nonlinear Evolution
Equations that. Change Type
ONE-DIMENSIONAL THERMOMECHANICAL
PHASE TRANSITIONS WITH NON-CONVEX
POTENTIALS OF GINZBURG-LANDAU TYPE
JéRGEN SPREKELS*
Abstract. In this paper we study the system of partial differential equations governing the nonlinear
‘thermomechanical processes in non-viscous, heat-conducting, one-dimensional solids. To allow for both
stress- and temperature-induced solid-solid phase transitions in the material, possibly accompanied by
hysterenin effects, a non-convex free energy of Ginzburg-Landau form is assumed. Results concerning the
‘well:posediness of the problem, as well as the numerical approximation and the optimal control of the
solutions, are presented in the paper, in particular in connection with the austenitic-martensitic phase
‘transitions in the so-called “shape memory alloys”.
Key words. phase transitions, non-convex potentials, Ginzburg-Landau theory, shape memory alloys,
hysteresis, conservation laws.
AMS(MOS) subject classifications. 35165, 25K60, 73U05, 73B30
1, Introduction. In this paper we consider thermomechanical processes in non-
viscous, one-dimensional heat-conducting solids of constant density p (assumed normalized
to unity) that are subjected to heating and loading. We think of metallic solids that do
not only respond to a change of the strain ¢ = uz (u stands for the displacement) by
an elastic stress ¢ = o(€), but also react to changes of the curvature of their metallic
lattices by a couple stress j: = (2). Thus, the corresponding free energy F is assumed in
Ginzburg-Landau form, i.e,
(1.1) F=F(6,€2,6),
where @ is the absolute temperature. In the framework of the Landau theory of phase tran-
sitions, the strain plays the role of an “order parameter”, whose actual value determines
what phase is prevailing in the material (see [3)).
Since we are interested in solid-solid phase transitions, driven by loading and/or heat-
ing, which are accompanied by hysteresis effects, we do not assume that F(¢,¢2,6) is a
convex function of the order parameter ¢ for all values of (cz,). A particularly interesting
'achbereich Bauwesen, Universitit-GHS-Essen, D-4300 asen 1, West Germany (visiting at IMA).
This work was supported by Deutsche Forschungsgemeinschaft (DFG), SPP “Anwendungsbezogene Opti-
micrung und Steuerung”.class of materials are the metallic alloys exhibiting the so-called “shape memory effect”.
Among those there are alloys like CuZn, CuSn,AuCuZn2,AgCd and, most important,
TiNi (so-called Nitinol). In these materials, the metallic lattice is deformed by shear, and
the assumption of a constant density is justified. The relation between shear stress and
shear strain (o—e-curves) of shape memory alloys exhibit a rather spectacular temperature-
dependent hysteretic behavior (see [2] for an account of the properties of shape memory
alloys):
a
Fig 1. Typical o —¢-curves in shape memory alloys, with tempera-
ture @ increasing from a) to c).
In addition, for sufficiently small shear stresses @ another hysteresis occurs in the
«—6-dingrome:
€
Fig 2. €~8 curves in shape memory alloys for different values of 0.On the microscopic scale, this hysteretic behaviour is ascribed to first-order stress-
induced (fg. 1a,b) or temperature-induced (fig. 2) phase transitions between different con-
figurations of the metallic lattice, namely the symmetric high-temperature phase “austen-
ite” (taken as reference configuration) and its two oppositively oriented sheared versions
termed “martensitic twins”, which prevail at low temperatures (cf., (6], {7]).
‘The simplest form for the free energy F which matches the experimental evidence given
by figs. 1,2 quite well and takes interfacial energies into account is given by (cf., (4, [5])
(1.2) F(e,€2,8) = —Cy@log(6/O2) + Cv0 + G+ 61(8 — O1)e? — pe! + nae? + : ay
where Cy denotes the specific heat, Gis some constant, 6; and @ are (positive) tempera-
tures and «1,9, &3,’/ are positive constants. A complete set of data for the alloy AuCuZnz
is given in {5]. Note that within the range of interesting temperatures, for 8 01, F is
not convex as function of €.
In the sequel, we assume F in the somewhat more general form (with positive «1, «2,7)
(1.3) F(¢,€2,8) = —Cy@ log(8/62) + Cv8 + 6 + wiOFi(c) + a Fa(e) + - é,
where F; and F, satisfy the hypothesis:
(Hl) Fi, rect
Fh(e) 2G lel-G, Ve ER, with positive constants %,72.
The dynamics of thermomechanical processes in a solid are governed by the conservation
Jaws of linear momentum, energy and mass. The latter may be ignored since p is constant
for the materials under consideration (we assume p = 1). The two others read
(14a) Ug Oet Hes =f,
(1.4) eet ge 06 Hear =9-
Here the involved quantities have their usual meanings, namely: o-elastic stress, p-couple
stress, u-displacement, f - density of loads, ¢ - density of internal energy, q - heat flux, g
-density of heat sources or sinks.
We have the constitutive relations
or OF pg OF
(1.5) o= 5°
and we assume the heat flux in the Fourier-form
(1.6) q
‘x6, , where x > 0 is the heat conductivity.
3Notice that (1.6) implies that the second principle of thermodynamics in form of the
Clausius-Duhem inequality is automatically satisfied.
Inserting (1.3), (1.5), (1.6) in the balance laws and assuming one-dimensional sample
of unit length, we obtain the system
(1.7a) te = (Kr OF (6) + 62FQ(6)), + Yteass = fy
(1.7) CO — Ki OF; (eer — KOz2 = 9,
(1.7e) e=us,
to be satisfied in the space-time cylinder 27, where T > 0, = (0,1), and, for t >
0, % = 2x (0,1).
In addition, we prescribe the initial and boundary conditions
7d) u(z,0) = uo(z), ue(z,0) = u(#), 6(¢,0) = @(z),2 € 9,
(1.7e) u(0,t) = uz2(0,t) = u(1,t) = use(1,t) = 0, #€ [0,7],
(up 9-(0,4) = 0, —n82(1,t) = B(O(L,t) ~ 6r(#)), #€ (0,7),
where 8 > 0 is a heat exchanige coefficient, and @p stands for the outside temperature at
ral
In the following sections we state some results concerning the well-posedness of the
system (1.7a-f), including a convergent numerical algorithm for its approximate solution.
To abbreviate the exposition, all constants in (1.7a-f) are assumed to equal unity; this will
have no bearing on the mathematical analysis.
2. Well-posedness. We consider (1.7a-f). In addition to (H1), we generally assume:
(H2) uy € (0) = {ue H4(Q)}u(0) = u(t) = 0 = w"(0) = w"'();
ur € HQ) H4(2); 6 € H2(M), bo(2)>0, Vee®
(H3) €5(0) = 0, 6r(0) = 60(1) + 4 65(1) > 0 (compatibility).
(4) f.g € (0,7; H*(9)), 6p € H4(0,7), where g(z,t) > 0 on Gr and Or(t) > 0
on [0,7].
‘We have the result:
THEOREM 2.1, Suppose (H1)-(H4) hold. Then (1.7a-f) has a unique solution (u,6)
which satisfies
(21a) we W(0,7; 17(0)) NW" (0,7; (9) 1 H*(Q)) NL” (0,7; H4(9)),
(2.1b) 6 H1(0,7;#*(Q)) NL? (0,7; #°(9)),
(21c) (2,1) >0, on Tr.Moreover, the operator (f,g,6r) + (u,8) maps bounded subsets of H1 (0,7; H*(9)) x
H'(0,T; H1(Q)) x M into bounded subsets of X x Y, where M := {z € |H1(0,7)| z(t) > 0
om [0,7], 2(0) = 80(2) + 04(1)} , X s= W2(0,7; L4(Q)) NWH(0,7; 41(2)N (2)
L°(0,T;H*(Q)) and Y := H*(0,7;H'(Q)) NL? (0,7; H°(Q)), respectively.
Proof. ‘The existence result is easily obtained by combining the Galerkin approximation
employed in the proof of Theorem 2.1 in [9] with the a priori estimates derived in the proof
of Theorem 2.1 in [12}; the uniqueness is a direct consequence of the subsequent Theorem
2.3. Finally, the boundedness of the mapping (f, ¢,6r) + (1,8) follows from the above-
mentioned a priori estimates. 0
A sharper existence result, with regards to the smoothness properties of the solution
(u,@), has been established in (12):
THEOREM 2.2. Suppose that, in addition to (H1)-(H4), the following assumptions on
the data of (1.7a-f) are satisfied:
(2.2) up € H(Q), u, € H°(2), H € H*(M),
fu € DOr), 9 €17(0,T;H7(Q)), Or € H?(0,7) .
Furthermore, suppose that 0 satisfies compatibility conditions of sufficiently high or-
der. Then (1.7a-f) has a unique classical solution (u,@), and all the partial derivatives
appearing in (1.7a-c) belong to the Hélder class C™*/2(Sp), for some a € (0,1).
Proof. See Theorem 2.1 in [12]. 0
We now derive a stability result with respect to the data (f,9,ér) which guarantees
the uniqueness of the solution (1,6).
‘THEOREM 2.3. Suppose the general hypotheses (H1), (H2) and 64(0) = 0 are satisfied.
We consider the variational problema
eee
(230) fulsrtoteey de furtayete.o)de— ff use dr
i id
+f | [RE con6¥e0 + uertee— fo] de dr=0,
a4
Vp € H"(0,7; (2) n27(0,7;H%(9)) , 0 0 such that
r
@4) (iment + lites(@)I? + [8(8)I7) + f f Gide dt
aa
exp
1€(0,7)
r
+ [20,8 & < CUlbeltsen + lela) + lean):
d
where 8 = 6) — 9), u = ul) = ul), 6p = A — 0, f = FO — Fg = gO) — g.
REMARKS.
1. Here (and throughout) we have omitted the arguments of the involved functions if
no confusion may arise.
2. |]. || denotes always the L*()-norm.
3. Obviously, any solution (u, 8) of (1.7a-f) with (2.1a,b) solves (2.9a-c); consequently,
the solution of (1.7a-f) is unique,
4. From the upcoming proof it will become evident that a corresponding stability
result holds with respect to the initial data uo, t,6o; we restrict ourselves to the
6data (f,9, 6r) as they are the natural candidates to serve as control variables if the
system is to be controlled from the outside.
Proof, Let e = u{?, i = 1,2, and ¢ = €(!) — ), In terms of the variables (u, 6)
introduced in the assertion, (2.3-c) can be rewritten as
(2.5a) Jutesnetenée f fruoe de de ff oeoee— Fels
a aa aa
t
IF or
“ff (F (69,0) =F (0, 0)) dz dr =0,
Vee MO,THO) NOTH), OSt O be given (to be specified later). We insert @ = tus in (2.6a), integrate
by parts and use Young's inequality to arrive at the estimate
:
@) 5 lea? +5 Nese(OP < 3 Baan) +O f h(a
i
' 12
+f {| (= (0M) — = (<),0) dzdr.
He
Now
> 2 (FE r,00y-F 0,90) = eae
+02 (FE) — FC) + (Jee
+ uF (6) — FE) 4+ FO uD
+ u20 (FYE) — FYE) + F(t.
7Consequently, invoking the mean value theorem,
(28) Sf (= (00) — z (0)
d :
o
2
de dr
:
0,02 > 0:
(2.16) NOC* Mccoy Sr 6c) H HOC) + exa0¢) |?
SHl6e(r)IP + Croll (DIP
Since 6,” € L*(0,7; L*()), this implies that
‘ t
(2.17) [OP Pe dads < J |O(r)Ilé~ all |Par
[ [otros]
: :
< Cob f f @dedr + Cy i f Paadt,
3a 3a
whence
'
(as) Il 0 sufficiently small, and invoking Gronwall’s
lemma, we have finally proved the assertion. 0)
3. Optimal Control. We now turn our interest to optimal control problems associ-
ated with the system (1.7a-f). It is of considerable interest in the technological application
of shape memory alloys to control the evolution of the austenitic-martensitic phase transi-
tions in the material; in this connection, a typical object is to influence the system via the
natural control variables f,g,6r in such a way, that » desired distribution of the phases
in the material is produced. Since the phase transitions are characterized by the order
parameter ¢, it is natural to use ¢ as the main variable in the cost functional. We consider
the following control problem:
(ap)
.
Minimize J(u, 6: f.9,8r) = J | Z1(2,t,ue(2,t),(2,), fest), o(2,t) ded
ta
t
+ fta(tGo()as f Za(emele 2), Oe,T))Ar,
a 4
subject to (1.7a-4) and the side condition (f,9,6r) € X, where K denotes some nonempty,
bounded, closed and convex subset of H? (0,7; H1(9)) x {g € H} (0,7; H1(2))|g(,t) > 0
on Mir} x Me
For Ly:
+ R, Ly: R? > Ry Ly: R? > R, we assume:
(H5) (i) L1, Lz, Ls are measurable with respect to the variables (2,#), resp. ¢, resp. 2, and
continuous with respect to the other variables.
(ii) Ly is convex with respect to f and g.
(iii) Lp is convex with respect to Op.
‘These assumptions are natural in the framework of optimal control; a typical form for
J would be
(a) ut, 85 ,98r) = Balle — Bellzacany + Ball? ~ Bacon)
+ Ballua(-,T) = a? + 6.|16(,7) ~ BI?
+ Asllflikacaz) + Bolla liZacaz) + Frl6rlli2¢0,2)>
with 8; > 0, but not all zero, and functions 7,8 € L?(r), #8 € L3(), representing the
desired strain and temperature distributions during the evolution and at t = T.
‘There holds
1THEOREM 3.1. Suppose (H1)-(H5) are true, then (CP) has a solution (w,3; 7,
Proof. Let {(fa,9ns6rn)}&X denote a minimizing sequence, and let (uin,9n) denote
the solution of (1.7a-f) associated with (fa, gn,6rjn),n € N. Since K is bounded, we may
sésume that
(3.2) In>F, weakly in H"(0,T;H*(Q)),
Jn, weakly in H*(0,7;H*(Q)),
rn > Br, weakly in H4(0,T).
‘Due to the weak closedness of the convex and closed set K,(7,9, Or) € XK. Let (3,8) denote
the associated solution of (1.7a-f). Now, owing to the boundedness of K and Theorem 2.1,
{(n/6n)}nen is @ bounded subset of X x Y. Therefore, we may assume that for some
(u,8) € X x ¥ there holds
Une > Uey
(8.38) a
uniformly on Qr,
Unjee > Utes
(3.3b) Unjex 7 Urey weakly in L?(Qr),
Unt > Uses
Unezes > Users,
as well as
Baz > Oxy
(3.3c) But + 4, weakly in L?(Qr).
On,2z > Pex
Passing to the limit ae n ~+ 00 in the equations (1.7a-f) shows that (1, @) solves (1.7a—
£) for the data (f,9,8r), ie, u = 1,0 = 6. Hence, (,6;f,9,5r) is admissible, and, in
view of (HS),
G4) Ji,
5,6r) < dim, inf J(un; Oni far 9ny9ryn)-
‘Thus (1,8; ,9, 6r) is a solution of (CP). 0
REMARKS.
5. The above way of arguing follows the lines of [10] where a related result was derived
for a much more restricted free energy F.
6. It is natural to look for necessary conditions of optimality for the optimal controls
of (CP). A corresponding result has not yet been derived.
7. The problem of the automatic self-regulation of the system via a fixed feedback
control regulating the boundary temperature 6r has been considered in {11}.
124, Numerical Approximation. In this section we follow the lines of [8]. We assume
the free energy in the special form (see (1.2))
ane oa_layleyle
(4.1) F(6sée,8) = Glog +643 02 FAs best a
Let
atoe-layle
(4.2) Fo(e8) = 5 OP —Get Ee.
Then, for e, # €2,
(43) Foless) — Folens) gees + 09) 4 W(e1,62),
ae 2
where Y(¢1,€2) is a polynomial of degree 5 in «1, €2-
We are going to construct a numerical scheme for the approximate solution of (1.7a-f).
To this end, we assume that (H1)-(H4) and (2.2) hold, so that Theorem 2.2 applies.
Now let K,N,M € N be chosen. We put h = Z, #2 = mh, 0< m< M, and
AM =H 0S SN.
Define
(4.4) Yw = {linear splines on [0,1] corresponding to the partition {2{")}, of {0,1]},
and let
(45)
= span{z1,.-.,2K})
where z; denotes the j-th eigenfunction of the eigenvalue problem
(4.6) 2!" = Az , in (0,1), 2(0) = 2"(0) = 0 = z"(1) = 2(1).
‘We introduce the projection operators
(4.7) Px = H*(0,1) — orthogonal projection onto Zx,
Qx = H?(0,1) ~ orthogonal projection onto Zx,
Ry = H*(0,1) — orthogonal projection onto Yy,
and the averages
En mt
(48) ime=]— f Hebd, ote)=F [sets
(m8 (nnn
ae
= fH)
myn
13‘We then consider the discrete problem
(Oun.x) Find u™ = Th oP a,6" = Th aryl”,
1 0,C, > 0, which do not depend on M,N, K, such that
for shr < + < Cy the discrete problem (Dy,w,x) has a solution which satisties
(4.108) @™(z)>0, Veet, 0