The value returned by fnormal is an expression equivalent to e under the assumption that all numeric values with magnitude less than epsilon may be considered to be zero.

In addition, all floats in e which remain nonzero are converted to floats with digits precision.

fnormal preserves numeric type and sign information as much as possible.  Thus, for example, fnormal(1e-20*I, 10) = 0.*I not 0. .  This ensures that branching behavior is generally not affected by fnormal.  Use simplify(expr, zero) to remove 0 real or imaginary parts of complex floating point numbers.

If e is a list, set, range, series, equation, or relation, then fnormal is applied recursively to the components of e.