next up previous notation contents
Next: 3.2.6 Piecewise Models Up: 3.2 Constant Interval Arithmetic Previous: 3.2.4 Constant Functions

3.2.5 Optimality

Consider the interval model tex2html_wrap_inline34387 of the function tex2html_wrap_inline34165 . The function g has many interval models; we will now define when the model tex2html_wrap_inline31397 is optimal.

A bound tex2html_wrap_inline34395 is optimal, for interval arithmetic, if no better upper bound exists:

math12429

The model tex2html_wrap_inline34397 returns optimal upper bounds if the upper bound is optimal for all tex2html_wrap_inline34383 :

math12438

We may now prove that the interval extension of g is optimal. Consider the upper bound, for argument j:

math12445

from the definition of interval extension. The only way tex2html_wrap_inline34405 could fail to be optimal is for there to be a better bound of g(j). This contradicts the definition of supremum; let the better bound be denoted as l,

math12452

or, equivalently:

math12457

but:

math12462

We now know that if g is differentiable over tex2html_wrap_inline34413 , then the upper bound given by tex2html_wrap_inline31397 is obtained by tex2html_wrap_inline34417 for some tex2html_wrap_inline34419 in j:

math12469

since j is closed. Lower bounds are handled similarly, and will be addressed in section gif.

Optimality can be defined without direct reference to the underlying function:

math12476

It is clear that tex2html_wrap_inline34425 since if tex2html_wrap_inline34427 then tex2html_wrap_inline34429 is clearly not optimal, so tex2html_wrap_inline31397 is not optimal. If tex2html_wrap_inline31397 is valid, then tex2html_wrap_inline34435 .


next up previous notation contents
Next: 3.2.6 Piecewise Models Up: 3.2 Constant Interval Arithmetic Previous: 3.2.4 Constant Functions
Jeff TupperMarch 1996