Linearizing convection terms in the Navier-Stokes equations

Standard approaches to dealing with the convection term in the Navier-Stokes equation involve either nonlinear Newton methods or Picard linearization. This paper focuses on Picard linearizations. Picard updates to the Navier-Stokes convection terms are accepted to be of the form ((u) over bar(N-1).del)(u) over bar(N). Current literature either ignores the possibility of, or recommends against, the ((u) over bar(N).del)(u) over bar(N-1) form. Why? Although this form is attractive due to its symmetry within a Finite Element discretization, it is not merely recommended against, it is wholly incorrect! This paper offers evidence to this effect in the form of analysis, numerical experiment and rationale.