15

I'm a bit confused on what exactly the meaning of a 'key' is in computer science. I understand key-value pairs, primary keys, etc... But I can't find a definition of what the term 'key' means by itself.

As far as I can tell it just means a piece of data. In CLRS, data associated with tree nodes are referred to as 'keys'. Data to search a hash table is called a 'key'. Is this what a 'key' is?

TheMax
  • 177
  • 1
  • 1
  • 5
  • 18
    There is not one specific technical definition. The use of the word is usually inspired by its normal, English definition, e.g. https://www.merriam-webster.com/dictionary/key Or rather I should say "definitions". In general, your expectation should be that there isn't a unifying technical definition for common English words that are used in multiple contexts even within a single field of study. – Derek Elkins left SE Apr 23 '19 at 19:52
  • 4
    It could even be one of those things on your keyboard :-) – jamesqf Apr 24 '19 at 03:50
  • It's actually the same in normal English - key figure = the primary person in the story, key piece of evidence = the primary evidence that lead to solving a case, key = the primary mechanism to unlock a door etc. It means "the main way to access something" in English. It's not specific to CS – slebetman Apr 24 '19 at 13:24
  • There are also "keys" in the cryptographic sense, which I would consider to be different from the data lookup examples you have mentioned. – 200_success Apr 24 '19 at 18:19
  • @slebetman While 'key' indeed has many uses in the English language, there are many usages that are have a precise definition that is very much specific to (a subfield of) CS. – Discrete lizard Apr 25 '19 at 07:29

2 Answers2

31

In the most general sense, a key is a piece of information required to retrieve some data. However, this meaning plays out differently depending on exactly what situation you're dealing with.

In the contexts you mention, a key is a unique identifier for the complete data used to retrieve it from some location in the structure. Each key is associated with only one item, so it can be used to find a particular set of data. The data structure will usually be organized in such a way that finding the key is much more efficient than a linear search through all of the data. Sometimes the key is actually part of the data and stored along with it (like primary keys in the database); other times, it is segregated from the data itself (like in a hash map). The data structure will also often perform extra processing on the key (and only the key) to support its efficient searching algorithm (such as in a hash map, the key is converted into a hash code, or a database will index the primary keys using a B-tree).

In cryptography, a key is used in a sense more akin to physical keys used on locks. They're pieces of data required to obtain the original from the encrypted data (to "unlock" the data, so to speak).

jpmc26
  • 428
  • 3
  • 6
  • 4
    To prevent possible confusion: in the book CLRS, keys are usually not considered to be unique, as they don't have to be for many data structures. – Discrete lizard Apr 23 '19 at 19:52
  • Then, in general, a key is data to navigate a data structure? That makes sense to me, like a physical key is used to retrieve something from a locked box. – TheMax Apr 23 '19 at 20:57
  • @TheMax I would not say that definition suits cryptography, as there's no "navigation" to be done. It does suit your list of examples, but I don't see it as a parallel to a physical key in those cases. – jpmc26 Apr 23 '19 at 21:08
  • @jpmc26 that description is spot on, consider bitwise XOR of a key against data, – mckenzm Apr 24 '19 at 06:37
  • On scales we see today, hashes used for synthetic keys may indeed not be unique, and may need tie breakers or compounding. – mckenzm Apr 24 '19 at 06:39
  • @mckenzm Collisions are a potential problem with all types of hashes, aren't they? Hash maps must deal with that possibility as part of their structure to function correctly. That's why the hash is not the key itself, but an additional processing step the hash map performs to optimize searches, but it still has to store the key. – jpmc26 Apr 24 '19 at 07:59
  • In your view of the key as a unique identifier, how is then a key different from a pointer to your data object? – Bananeen Dec 16 '20 at 18:28
13

A key in the context of data structures (such as in the book CLRS) is a value (often an integer) that is used to identify a certain component of a data-structure. Often, keys determine how the underlying data is stored or manipulated. For example, in binary search trees we have that for every node, the key of that node is larger than the keys in the left sub-tree and smaller than those in the right subtree. This property makes it easier to search for a given key (or determine there is no node with such a key).

In practice, our 'actual' data is often not a key, but something larger and more relevant that a single number. This data is called satellite data and can be mostly ignored when dealing with manipulations on data structures, as long as the satellite data moves whenever the key gets moved (otherwise, you lose track of your data).

The concept of a key is similar in the context of databases, but there it is often required that a key is unique. A primary key has to be unique, for example. This requirement is often nessecary in the context of data-structures, but is sometimes made for simplicity.

In cryptography, a key usually refers to an (often secret, but not always!) parameter that is needed to encrypt or decrypt with a given en- or decryption algorithm. The keys used to encrypt and decrypt have to be 'related' (in symmetric cryptography, the need to be the same) for the process to encryption or decryption to be successful.

Discrete lizard
  • 8,248
  • 3
  • 24
  • 53
  • What is the difference between satellite data and keys then? From what I understand, satellite data is data organized by the data structure that is not part of the actual structure. So can I say a keys and satellite data are both data in the structure, but keys are part of the structure and satellite data is not? – TheMax Apr 23 '19 at 19:18
  • 1
    @TheMax In a way, yes. The precise content of the satellite data is irrelevant for the operations on the data-structure (but likely relevant for the application using the data structure). This decoupling of key and data makes it easier for to design efficient data-structures. – Discrete lizard Apr 23 '19 at 19:24
  • Great answer! Can you clarify though in what sense is then a key different from a pointer (in the context of data structures)? Both allow us to identify the object with the relevant satellite data. Is it also correct to say that a key is in some sense a generalized index in the sense that it it is a value with some order structure (defined by the data structure one is considering) generalizing how indices are linearly ordered in an array? – Bananeen Dec 16 '20 at 18:25
  • 1
    @Bananeen You can see a pointer as a type of key, that points to a specific (possibly virtual) memory location. But a key doesn't have to be a description of a specific memory location, it could be an IP address, geographic coordinates, a book title, anything, really. Which is the point of the terminology: when designing an algorithm, we want to abstract such implementation details away. – Discrete lizard Dec 16 '20 at 19:04
  • 1
    @Bananeen A key can often be used to define an ordering if the set of keys itself has a meaningful order, such as when your keys are integers. But if your keys are e.g. geographic coordinates, what this order should will be less clear. Sure, you can think of a way to order geographic coordinates, but you can also think of an order based on other information in the satellite data. I think a key doesn't seem to have a special role here in general, although it can and often is used to define an ordering in certain contexts. – Discrete lizard Dec 16 '20 at 19:04
  • 1
    @Discretelizard Thank you for the clarification. So building on what you said "it could be an IP address, geographic coordinates, a book title, anything, really.", a key is just a chosen attribute of the objects we are considering that is going to be important for the algorithm at hand. Other, non-important attributes we treat under the hood of "satellite data". Is it a fair summary? – Bananeen Dec 16 '20 at 20:23