next up previous notation contents
Next: 3.3.30 Floating Point Up: 3.3 Linear Interval Arithmetic Previous: 3.3.28 Examples with a Binary

3.3.29 Concave Up, Down Functions

There is another way to approach evaluating tex2html_wrap_inline31401 , when g is neither concave up nor concave down. We consider g such that tex2html_wrap_inline37297 . Restricting our attention to continuous g, we may find an upper bound without splitting g into two parts. Consider the following chart, a tex2html_wrap_inline35653 chart where tex2html_wrap_inline37305 and tex2html_wrap_inline37307 .

figure24336

First, bounds of tex2html_wrap_inline37327 and tex2html_wrap_inline37329 are found; tex2html_wrap_inline37331 , tex2html_wrap_inline37333 . Let G' be formed by connecting the left endpoint of tex2html_wrap_inline37337 with the left endpoint of tex2html_wrap_inline37339 . If this line extends so that it overlaps tex2html_wrap_inline37339 , then this may be taken as an upper bound of both sections. A lower bound would be found for the above example with the procedures outlined earier.

Another approach is to find a quadratic upper bound and then produce a linear upper bound of the quadratic upper bound. This may be done if tex2html_wrap_inline37343 .

figure24447


next up previous notation contents
Next: 3.3.30 Floating Point Up: 3.3 Linear Interval Arithmetic Previous: 3.3.28 Examples with a Binary
Jeff TupperMarch 1996