Optimally designed structures under eigenvalue constraints (natural fr
equency or buckling loads) are likely to display multiple eigenvalues.
This introduces singularity of eigenvalue derivatives with respect to
the design vector, which does not allow use of the Kuhn-Tucker condit
ions. The paper presents the optimality criteria which takes that sing
ularity into account and, in contrast to commonly used Kuhn-Tucker con
ditions, they are valid for multiple eigenvalues (the form of that cri
teria is more general than the one derived from the Kuhn-Tucker condit
ions). An algorithm is presented, which applies those criteria and als
o uses a new method of computation of Lagrange multipliers. The algori
thm allows for an arbitrary number of eigenvectors to be considered in
optimality criteria, and the actual modality of the problem is determ
ined automatically. Examples of optimization of truss structures model
led by finite element method are included.