fhom(z) = f(z).
If we make the above example only a little more complex considering also convex next-
to-nearest-neighbour (NNN) interactions; i.e., fn= 0 for all n≥3, with f1and f2convex
functions, we loose this exact characterization of the minimal arrays. However, the discrepancy
between uk
jand zj decays fast away from the endpoints j= 0 and j=kof the array.
A slight adjustment of the boundary-value problems, say by imposing additional boundary
conditions u1=zand uk−1=z(k−1) (which do not influence the asymptotic value of
the minima in (1.2)) reestablishes the affine interpolations uk
j=zj as minimizers, so that
fhom(z) = 2(f1(z) + f2(2z)). In this case the classical Cauchy-Born rule is applicable, given
that we modify boundary conditions in the ‘cell’ problem. Note that this analysis extends to
any sufficiently fast decaying set of convex potentials fn, giving fhom(z) = 2 P∞
n=1 fn(nz).
Even if we abandon the convex setting, we may still easily describe the behavior of min-
imum problems in (1.2) in the case of nearest-neighbor interaction, with f1=f. It can be
shown that fhom in (1.2) is given by the convexification f∗∗ of the NN potential [25]. However
the classical Cauchy-Born rule in this case has to be properly generalized. Suppose, for in-
stance, that the potential fhas a double-well form. In this case the relaxation points towards
configurations containing mixtures of the two energy wells. Since in this setting there are no
obstacles to simple mixing, the relaxation strategy providing fhom is straightforward. Indeed,
for each zthere exist z1,z2,θ∈[0,1] such that f∗∗(z) = θf(z1) + (1 −θ)f(z2). Hence, we
can construct a function uz:Z→Rwith uz
i−uz
i−1∈ {z1, z2},uz
0= 0 and |uz
i−iz| ≤ C.
Such uzmay be chosen periodic, if θis rational, or quasiperiodic (loosely speaking, as the
trace on Zof a periodic function with an irrational period) otherwise. In both cases we obtain
‘local’ minimizers with ‘global’ properties which allows one to talk about the applicability of
the GCB rule.
The situation is more complex in the case when non-convexity is combined with frustrated
(incompatible) interactions. To show this effect in the simplest setting it is sufficient to
account for nearest-neighbor and next-to-nearest-neighbor interactions only and we make the
simplest nontrivial choice by assuming that f1is a ‘double-well’ potential and that f2is a
convex potential. In this case the homogenized potential fhom is also known explicitly [24, 77].
Its domain can be subdivided in three zones: two zones of ‘convexity’ where minimizers are
trivial (as for convex potentials) and a zone where (approximate) minimizers in (1.2) are
two-periodic functions with uz
i−uz
i−1∈ {z1, z2}and z1+z2=z(in a sense, a constrained
non-convex case as above). Hence, in these three zones we have minimizers with a ‘global’
form because the macroscopic energy can be obtained by solving elementary ‘cell’ problems.
One can say that in the two zones of ‘convexity’ the classical CB rule is applicable. In
the ‘two-periodic’ third zone we see that the homogeneity of the minimizers is lost but an
appropriately augmented GCB rule still holds. For the remaining values of zno ‘local’ GCB
rule is applicable since in those cases the unique (up to reflections) minimizer is a ‘two-phase’
configuration with affine and two-periodic minimizers coexisting while being separated by a
single ‘interface’ [22]. The frustration (incompatibility) manifests itself in this case through
the impossibility of the penalty-free accommodation of next-to-nearest interactions across such
4