Feng contributed to several fields in mathematics. Before 1957 he mainly worked on
pure mathematics, specially on
topological groups,
Lie groups and generalized function theory. From 1957 he began studying
applied mathematics and
computational mathematics. He made a series of discoveries in computational mathematics. In the later 1950s and early 1960s, based on the computations of dam constructions, Feng proposed a systematic
numerical technique for solving
partial differential equations. The method was called the
Finite difference method based on variation principles (). This method was also independently invented in the West, and is more widely known as the
finite element method. It is now considered that the invention of the finite element method is a milestone of computational mathematics. In the 1970s Feng developed
embedding theories in discontinuous finite element space, and generalized classical theory on
elliptic partial differential equations to various dimensional combinations, which provided a mathematical foundation for elastic composite structures. He also worked on reducing elliptic PDEs to
boundary integral equations, which led to the development of the
natural boundary element method, now regarded as one of three main
boundary element methods. Since 1978 he had given lectures and seminars on finite elements and natural boundary elements in more than ten universities and institutes in France, Italy, Japan and United States. From 1984 Feng changed his research field from elliptic PDEs to
dynamical systems such as
Hamiltonian systems and
wave equations. He proposed
symplectic algorithms for Hamiltonian systems. Such algorithms preserve the
symplectic geometric structure of Hamiltonian systems. He led a research group which worked on symplectic algorithms for solving Hamiltonian systems with finite and infinite dimensions, and also on dynamical systems with
Lie algebraic structures, such as
contact systems and
source-free systems. Since these algorithms make use of the corresponding geometry and the underlying Lie algebras and Lie groups, they are superior to conventional algorithms in long term tracking and qualitative simulation in many practical applications, such as
celestial mechanics and
molecular dynamics. == References ==