CGQL

where the difference [([T.sup.j+d.sub.-1]).sub.11] - [([T.sup.j.sub.-1]).sub.11] is computed by the algorithm CGQL from [16],

It is important to understand that the additional rounding errors in computing _j;d _j;d, [[upsilon].sub.j,d] and [v.sub.j,d] from the given formulas (the algorithm CGQL and (4.7)-(4.9)) do not affect significantly the values of the computed bounds and do not represent a problem.

11.1 demonstrates, that the estimates [[eta].sub.j,d].sup.1/2] j;d (computed by the algorithm CGQL [16], dotted line), [[upsilon].sub.j,d].sup.1 /2] (dashed line) and [v.sub.j,d].sup.1/2] (dash-dotted line) give in the presence of rounding errors similar results; all the lines essentially coincide until [[parallel]x - [x.sub.j].sub.A] (solid line) reaches its ultimate attainable accuracy level.