# 287: NP-Complete

 NP-Complete Title text: General solutions get you a 50% tip.

## Explanation

Another entry in the “My Hobby” series of cartoons; here, Cueball is embedding NP-complete problems in restaurant orders. Specifically, he is ordering appetizers not by explicitly stating the names of the appetizers, but by the total price of the chosen appetizers. This is a (simplified) example of the knapsack problem. The knapsack (or rucksack) problem is a problem in combinatorial optimization, as follows: if you have a knapsack (or backpack, or rucksack, as your regional desire to call it may be) which can hold a specific amount of weight, and you have a set of items, each with its own assigned value and weight, can you select items to put into the knapsack so that (1) the weight does not exceed the capacity of the knapsack, and (2) the combined value of all the items is maximized. (The example in this cartoon is simplified, because we are not given the corresponding “value” of each item – like we would be if, for example, Cueball and his two friends assigned a perceived value of 7 on a scale of 1 to 10 for getting hot wings, but only a 2 out of 10 for a side salad. Another way to introduce complexity into the problem would be if the patrons were considering the calorie values of each appetizer on the menu.)

The knapsack problem is, as indicated by the title of the strip, NP-complete. In computational complexity theory, NP stands for “nondeterministic polynomial time.” Basically, for an NP-complete problem, there is no efficient way to find a solution, but it is relatively easy to verify that a solution works. Conceptually, for the problem posed in this comic, the most straightforward way to find a solution (not even knowing for certain if there is one) is to methodically start by first listing all the (6) ways of choosing one appetizer, and their totals (which of course are the prices of each appetizer), then list all the (21) ways of choosing two appetizers (allowing the same appetizer to be chosen twice, since we are assuming that the chef can create more than one sampler platter), and then list all the (56) ways of choosing three appetizers, and so forth. As any combination of eight appetizers would be more than \$15.05, the process need not extend beyond listing all the (1715) ways of choosing seven appetizers.

The image text refers to the fact that NP-complete problems have no known general solution. (A general solution would be the optimal solution to this problem generalized to allow any prices for the appetizers.) If the waiter can find an efficient general solution to this, or any NP-complete problem, he will have solved one of the most famous problems in computer science, whether or not P=NP. In other words, if every problem whose solution can be quickly verified by a computer can also be quickly solved by a computer. This problem is one of the seven Millennium Prize Problems stated by the Clay Institute in 2000, six of which (including whether P=NP) are still unsolved, for which a correct solution is worth US\$1,000,000, so a 50% tip is slightly less than fair.

Another famous NP-complete problem is the Travelling Salesman problem (also see 399: Travelling Salesman Problem), also mentioned in the strip. An example: suppose a traveling salesman has to visit ten cities and then return home, and it is possible to directly travel from any city to any other city (by plane, for example.) What is the minimum distance (or time, or cost) necessary for the trip? Like the knapsack problem, it is (usually) difficult to find the best solution without trying a very large number of combinations. Cueball mentions this problem after the waiter states he has to tend to six other tables.

A film reference is embedded in the menu in the first panel: the restaurant is called “Chotchkies,” a fictional restaurant featured in the film Office Space. (In that film, the character Joanna, played by Jennifer Aniston, quits her job at Chotchkies, a typical family-oriented chain restaurant, over their policy that she wear a large number of “flair” items – tacky pins, buttons, or other adornments to a worker’s uniform which can often be seen on waiters and waitresses at chain family restaurants, as well as those who work at movie theaters or large retail chain stores.)

## Solution

There are exactly two solutions to the problem posed in the comic strip, combinations of appetizers which total \$15.05:

• seven mixed fruit orders
• a combination of two orders of hot wings, one order of mixed fruit, and one sampler plate

## Transcript

My Hobby:
Embedding NP-Complete problems in restaurant orders
Chotchkies Retaurant
Appetizers
Mixed Fruit 2.15
French Fries 2.75
Hot Wings 3.55
Mozzarella Sticks 4.20
Sampler Plate 5.80
[Three people sit at a table. One man at the table is ordering from a waiter]
Man at table: We'd like exactly \$15.05 worth of appetizers, please.
Waiter: ... Exactly? Uhh ...
Man at table: Here, these papers on the knapsack problem might help you out.
Waiter: Listen, I have six other tables to get to -
Man at table: - As fast as possible, of course. Want something on traveling salesman?

# Discussion

Unique deciphering requires unique pricetags

Shame this only works in restaurants that price all their appetizers differently. Davidy22 (talk) 03:18, 13 October 2012 (UTC)

Not necessarily because the NP-problem allows for any equivocally competing sum certifying how the total can be reached. Shared pricetags as well as a nonpositive would add degrees of freedom and make it impossible to rule out surprise deliveries even through exponential pretesting. Unless the waiter is running into the exponential worst case, the six waiting tables can be attended to immediately upon finding the first feasible combination: Roman Czyborra (talk) 15:44, 15 May 2014 (UTC)
Trivial solution first found

I have a hunch that the seven fruit cups are pretty intentional as the first item on the menu and the simplest solution possible. I was about to write a script to solve the problem through random selections and was going to optimize for speed by limiting the maximum times an item could be order to floor(15.05/price). Thus, one could order up to 2 sample plates, 3 moz sticks, 5 of the hot wings/side salad/french fries or 7 fruit cups without going over budget. (side note: you can always with these prices squeeze in a fruit cup with the exception of the 7 fruit cups). I found the "trivial" solution on the first step of the "preliminary" work for that script and then took a catnap. Of course, since the nontrivial solution involves the same item as the trivial solution, one could just pick a number, multiply by that number, subtract one unit, and pick two other items, whose prices were not set yet, and adjust their prices to add up accordingly just to ensure both trivial and nontrivial solutions lest anyone actually write a program to solve the problem through brute force as oppose to through wit. Why seed? Because to not have a nontrivial solution would be so much like Blackhat. Note to self: try this sometime in the real world using a real menu. Katya (talk) 02:17, 23 November 2012 (UTC)

Traveling Salesman Problem

Note: Traveling Salesman Problem might be mentioned also because both this problem and the Knapsack problem to be solved belong to set of NP-complete problems; a Knapsack problem can be transformed in polynomial time to Traveling Salesman Problem, and solution of Traveling Salesman Problem can be transformed in polynomial time to Knapsack problem solution. --JakubNarebski (talk) 16:00, 11 December 2012 (UTC)

Yes, indeed! I think both meanings are intended to fully get the joke. The `TSP:={(n,d,M)∈ℕ×({0…n}²→ℕ)×ℕ|∃c∈{1…n}ⁿ:{1…n}=⋃{cₙ|n∈{1…n}}∧∑{d(cₙ,c₍ₙ₊₁₎)|n∈{0…n}}<M}` can both help to timely attend to the six waiting tables and to reduce the `ORDERSUM:={(a,b)∈ℕ*×ℕ|∃c∈ℕ*:∑{cₙaₙ|n∈ℕ}=b}` problem to. Plus, the "as fast as possible" pun seems to allude to the again six ridiculous inputs any trained human will rearrange to a near-exact solution quicker than they are entered into a computer who can quickly exhaust this tiny search space for an exact solution: Roman Czyborra (talk) 15:44, 15 May 2014 (UTC)
Trivial solution was not intended

In an interview with the Mathematical Association of America Randall said that the trivial answer to this problem was a mistake. Xrays Knock Charms Down (talk) 03:00, 6 May 2013 (UTC)

I added this very interesting info to the explanation - at first as a trivia, but then I realized that it would not be seen by everyone - as you often do not read below the transcript. Why would you, you do not need to see what was in the comic again... So I moved it up to the solution part, because to me it is a very important fact about this comic. An error by Randall... But Dgbrt keeps moving this info away from the solution. I have understood now that the trivia should be below the transcript - although I cannot see why this should be so - as I have just described. But who says that this info should be a trivia item? It was I who put it there (by mistake?) at first. I will try not to start an editing fight here, but still think there should at least be a mention in the explanation that it was a mistake - in case you do not realize there is a trivia section below. I have used this page a lot lately, and had not found out before, that it was always below. There is not that many pages with trivia sections Kynde (talk) 11:02, 10 March 2014 (UTC)
Cool reference, thanks! Roman Czyborra (talk) 15:44, 15 May 2014 (UTC)
Complex solution found in a second

I was bored and tried to find a solution for fun. I found the more complex one quite fast by chance. It was actually the second combination I tried. I did not realize you could just add seven fruit cups because I was so set on starting with the sampler plate. Now I am not sure if I should be glad, because I was so lucky, or annoyed that my fight-the-boredom-idea did not work out, or even more annoyed that I never have that kind of luck in the lab where I could really use it for finding the one thing out of a thousand possible causes for "why-does-my-experiment-not-work" which actually will give me some usable data. 84.56.77.11

Not the knapsack problem

This explanation is thorough, and I like being thorough, but it seems to be a bit of overkill. I copy-edited it a bit, but I have a couple qualms. This is not really the knapsack problem, as it does not attach values to the items (as mentioned). It is more of a subset sum problem, which admittedly could be considered a variant of the knapsack problem. Secondly, I don't see why we need to go into detail about the movie Office Space. --Quicksilver (talk) 18:34, 22 August 2013 (UTC)

I did some clean-ups, but the the "In computational complexity theory" still needs a review.--Dgbrt (talk) 20:19, 22 August 2013 (UTC)
The Wikipedia article on Karp's 21 NP-complete problems hints that Karp originally defined `KNAPSACK:={(a,b)∈ℤ*×ℤ|∃c∈𝔹*:∑{cₙaₙ|n∈ℕ}=b}` closer to today's shape of `SUBSETSUM:={Z⊂ℤ|∃s⊆Z:∑s=0∧s≠∅}` than that of the Unbounded Knapsack Problem `UKP:={(v,w,V,W)∈ℤ*×ℤ*×ℤ×ℤ|∃c∈ℕ*:{∑{cₙvₙ|n∈ℕ},∑{cₙwₙ|n∈ℕ}}⊆{V…W}}` with the former via `Z:={b,-a₁…-aₙ,-2a₁…-2aₙ,…}` and the latter via `(v,w,V,W):=(a,a,b,b)` coming close enough to what we really need here, namely `ORDERSUM:={(a,b)∈ℕ*×ℕ|∃c∈ℕ*:∑{cₙaₙ|n∈ℕ}=b}`. So Randall did hit it bull's eye after all! Roman Czyborra (talk) 15:44, 15 May 2014 (UTC)
NP Food

Inspired by this comic, somebody has actually created an ordering site which tries to give you an order from a restaurant in your area (US only I think) totalling a specific amount NP Food. Worth including above? -- Copito (talk) 20:43, 8 November 2013 (UTC)

That site doesn't work for me. —TobyBartels (talk) 10:07, 19 November 2013 (UTC)

I do get more than nothing: a redirect to the HTTPS port whose certificate is signed only to .np-food.com without WWW and whose HTML and PNG and JS suggest that either solutions for San Francisco, Austin, Saint Louis, Miami, and New York menues have been memoized and that you may order by entering your credit card credentials or that only fools wait for a computer to calculate an NP-hard problem on too large a search space. Roman Czyborra (talk) 15:44, 15 May 2014 (UTC)
Exhaustive Solution

Roman Czyborra did post this at the explain:

The Solution

… can be calculated as

```let totaling total menu = if total == 0 then [[]]
else if total < 0 || null menu then []
else totaling total (tail menu) ++ map (
in totaling 1505 [215,275,335,355,420,580]
== [[215,355,355,580],[215,215,215,215,215,215,215]]
```

I don't think this is a helpful explain. --Dgbrt (talk) 19:11, 14 May 2014 (UTC)

Yes, I did. Because I did think it was helpful. Not just because an (effective if not efficient) general solution earns you a 50% on \$15.05 tip.

Moreover to demonstrate that and how a complete search finds those two solutions.

And that the search tree can branch exponentially with each additional menu item. Or with additional dollar bills to be spent. Notwithstanding that any constructive proof of NP=P would let us replace this straightforward bad NP-implementation with an equivalent better P-implementation. Before Donald Knuth coined the name NP-Complete, the class was suggested to be named PET for the (Probably(while NP?P)|(Proven(if NP>P)|Previously(if NP=P))) Exponential Time pet problems.

What is so confusing about the calculation? The whole cent amounts instead of dollar floats? My naming of variables? Should `totaling` be renamed to `solutions` or `orders`? Or `menu` to `menu_items` or `appetizers` or `pricetags`? ``` ```

``` type Cents = Int orders :: [Cents] -> Cents -> [ [Cents] ] orders menu total = total == 0 | [ [] ] menu == [] | [] total < 0 | [] total > 0 | orders (tail menu) total ++ map ( head menu :) orders menu (total - head menu) orders [215,275,335,355,420,580] 1505 == [[215,355,355,580],[215,215,215,215,215,215,215]] calls menu total = if null menu || total < 1 then 1 else 1 + calls (tail menu) total + calls menu (total - head menu) calls [] 1505 == 1 calls [580] 1505 == 7 calls [420,580] 1505 == 25 calls [355,420,580] 1505 == 73 calls [335,355,420,580] 1505 == 181 calls [275,335,355,420,580] 1505 == 437 calls [215,275,335,355,420,580] 1505 == 1153 ```

Or is it the committee language Haskell that is causing problems? What other well-defined language would you formulate a general solution in?

Discussing all of this is helpful. Leaving a "Thus" result without its afferent reasoning (and its deleted heading) is not, is it? Cheers: Roman Czyborra (talk) 15:44, 15 May 2014 (UTC)

Please let's keep this code at the discussion page. No common reader would understand; the explain is not only for programmers. I'm a programmer, knowing many languages like BASIC, Pascal, C, C++, Java, Bash, Perl... also HTML, JavaScript... RPG, Databases and SQL... and much more. And if you like to buy an IBM Power 8 I can tell you the proper configuration for your needs.
But these details are not helpful to explain the comic. There is math that has to be explained. Findings on program codes do even not belong to a trivia section. Nevertheless it seems I have to take a closer look on Haskell, which is not used by many people. --Dgbrt (talk) 21:22, 15 May 2014 (UTC)

A 50% tip on a \$ 15.05 order is not possible, is it? --108.162.231.186 21:08, 1 November 2014 (UTC)

If I were the waiter my response would, at best, be "I'll come back when you're ready to order". At worse it would probably involve burns. -Pennpenn 108.162.250.162 04:27, 7 October 2015 (UTC)

---Easiest response: "Excellent, Sir. I'll raise the price of the french fries to \$15.05 - Ruffy314 (talk) 18:19, 21 November 2015 (UTC)