# What is archimedean weighting?

 2 In the context of optimization algorithms, what is archimedean weighting? Especially examples are welcome :) (Thanks to Paul for bringing this keyword to my attention) asked 18 Feb '13, 08:56 Geoffrey De ... ♦ 3.3k●2●15●50 accept rate: 6% ...as found in this question? (18 Feb '13, 09:59) fbahr ♦ @fbahr yes, for example as found there, but doesn't have to relate to multi-objective optimization necessarily. I am looking for a clear, unambiguous example and/or definition. (18 Feb '13, 10:05) Geoffrey De ... ♦ 1 See here for another note from Paul. To me it seems more like a naming convention. (18 Feb '13, 12:33) Ehsan ♦

 4 Here, afaiu, @Paul refers to weighted goal programming, in which weights are assigned to deviations of objectives from their perspective goal. Let's say, you have a program: \begin{align} f_1(x) &\ge v_1\\ f_2(x) &= v_2\\ f_3(x) &\le v_3\\ x &\in X \end{align} Then, its "Archimedian" formulation is: \min \alpha_{s_1^-} \times s_1^- + \alpha_{s_2^-} \times s_2^- + \alpha_{s_2^+} \times s_2^+ + \alpha_{s_3^-} \times s_3^+\\ \begin{align} f_1(x) &+ s_1^- & &\ge v_1\\ f_2(x) &+ s_2^- &- s_2^+ &= v_2\\ f_3(x) & &- s_3^+ &\le v_3\\ s &\ge 0, & x &\in X\\ \alpha_i & > 0, & \sum \alpha_i &= 1 \end{align} answered 18 Feb '13, 10:06 fbahr ♦ 4.4k●5●15 accept rate: 12% Does archimedean weighting imply the use of deviations of objectives from their perspective goal? Stated differently, does it imply that if v1 = 10 that a solution with f1(x) = 5 should have a worse weight than another solution with f1(x) = 8? (18 Feb '13, 10:15) Geoffrey De ... ♦ Err, I failed to put the most important part of the "Archimedian" formulation in my answer above: in weighted goal programming, the $$\alpha_i$$'s (generally) can take any value ($$> 0$$); a weighting is called "Archimedian" iff (also) $$\sum \alpha_i = 1$$. (18 Feb '13, 10:31) fbahr ♦ So, whether a solution with $$f_1(x) = 5$$ has a worse weight than an(y )other solution with $$f_1(x) = 8$$ may depend on your choice of $$\alpha_1$$ – but since $$\alpha_1$$ is a positive constant, the "penalty" for having $$f_1(x) = 5$$ in your solution is always higher than for $$f_1(x) = 8$$ [may be over-compensated by $$f_2, f_3$$, though]. (18 Feb '13, 10:48) fbahr ♦ 2 For what little it's worth, I usually see "Archimedean" used in the context of goal programming (which implies weighting deviation variables), but as far as I know the term can also be applied to taking a weighted linear combination of objective values to create a single criterion function. Non-archimedean weighting, in the context of goal programming, appears to be synonymous with "preemptive priorities". (18 Feb '13, 18:12) Paul Rubin ♦ 1 @Paul: As I told you before (and tell you again) - things would be so much easier for us if we had something like "Paul's Compendium of Q&As in the World of OR". Whenever a "new" question pops up, chances are that you already answered them somewhere, somewhere, seomewhy (on sci.op-research, IBM's opt. forums, etc.pp)! > sudo write-it! ;-) (19 Feb '13, 04:02) fbahr ♦ @fbahr: A couple of things to consider: (a) I can't recall when or where I've answered things, so it would be hard to assemble a compendium; (b) not all my answers are correct (and hence not all bear repeating). In fact, I can't swear that a majority of my answers have been correct. I learned along time ago that saying something with an air of certainty often trumps saying the correct thing. :-) (20 Feb '13, 11:15) Paul Rubin ♦ showing 5 of 6 show 1 more comments
 toggle preview community wiki

By Email:

Markdown Basics

• *italic* or _italic_
• **bold** or __bold__
• image?![alt text](/path/img.jpg "Title")
• numbered list: 1. Foo 2. Bar
• to add a line break simply add two spaces to where you would like the new line to be.
• basic HTML tags are also supported

Tags: