Simulations of magnetic and magnetostrictive behavior based on microma
gnetic theory exhibit hysteresis. These systems have a highly nonlinea
r character involving both short range anisotropy and elastic fields a
nd dispersive demagnetization fields. Hysteresis occurs even in the ab
sence of an imposed dynamical mechanism, for example, a Landau-Lifshit
z-Gilbert dissipative equation for the magnetic moment, and is symptom
atic of the way the system navigates a path through local minima of it
s energy space. It is not sensitive to the particular method: We imple
ment continuation based on the conjugate gradient method, although the
same results were obtained by, for example, a Newton's method. The ph
enomenon is robust: Computational experiments confirm that the shape o
f the loop is invariant over several decades of mesh refinement. Our e
xperience has led us to hold that optimization procedures have the pro
pensity to become marooned at local extrema when applied to nonconvex
situations and that this presents a fundamental challenge to analysis.
Understanding and controlling such phenomena present the opportunity
to develop predictive tools and diagnostics. For example, since the en
ergy picture is mesh-independent, computing on a fairly coarse grid su
ffices to establish its character.