A Proof-of-Work (PoW) system (or protocol, or function) is a consensus mechanism. It deters denial of service attacks and other service abuses such as spam on a network by requiring some work from the service requester, usually meaning processing time by a computer. The concept was invented by Cynthia Dwork and Moni Naor as presented in a 1993 journal article.^[1] The term "Proof of Work" or PoW was first coined and formalized in a 1999 paper by Markus Jakobsson and Ari Juels.^[2]

A key feature of these schemes is their asymmetry: the work must be moderately hard (yet feasible) on the requester side but easy to check for the service provider. This idea is also known as a CPU cost function, client puzzle, computational puzzle or CPU pricing function. It is distinct from a CAPTCHA, which is intended for a human to solve quickly, rather than a computer.

One popular system, used in Hashcash, uses partial hash inversions to prove that work was done, as a good-will token to send an e-mail. For instance the following header represents about 252 hash computations to send a message to calvin@comics.net on January 19, 2038:

It is verified with a single computation by checking that the SHA-1 hash of the stamp (omit the header name X-Hashcash: including the colon and any amount of whitespace following it up to the digit '1') begins with 52 binary zeros, that is 13 hexadecimal zeros: ^[]

Whether PoW systems can actually solve a particular denial-of-service issue such as the spam problem is subject to debate;^[3][4] the system must make sending spam emails obtrusively unproductive for the spammer, but should also not prevent legitimate users from sending their messages. In other words, a genuine user should not encounter any difficulties when sending an email, but an email spammer would have to expend a considerable amount of computing power to send out many emails at once. Proof-of-work systems are being used as a primitive by other more complex cryptographic systems such as bitcoin which uses a system similar to Hashcash.

There are two classes of proof-of-work protocols.

Known-solution protocols tend to have slightly lower variance than unbounded probabilistic protocols because the variance of a rectangular distribution is lower than the variance of a Poisson distribution (with the same mean). A generic technique for reducing variance is to use multiple independent sub-challenges, as the average of multiple samples will have a lower variance.

There are also fixed-cost functions such as the time-lock puzzle.

Moreover, the underlying functions used by these schemes may be:

Finally, some PoW systems offer shortcut computations that allow participants who know a secret, typically a private key, to generate cheap POWs. The rationale is that mailing-list holders may generate stamps for every recipient without incurring a high cost. Whether such a feature is desirable depends on the usage scenario.

Here is a list of known proof-of-work functions:

Computer scientist Hal Finney built on the proof-of-work idea, yielding a system that exploited reusable proof of work ("RPOW").^[18] The idea of making proofs-of-work reusable for some practical purpose had already been established in 1999.^[2] Finney's purpose for RPoW was as token money. Just as a gold coin's value is thought to be underpinned by the value of the raw gold needed to make it, the value of an RPoW token is guaranteed by the value of the real-world resources required to 'mint' a PoW token. In Finney's version of RPoW, the PoW token is a piece of Hashcash.

A website can demand a PoW token in exchange for service. Requiring a PoW token from users would inhibit frivolous or excessive use of the service, sparing the service's underlying resources, such as bandwidth to the Internet, computation, disk space, electricity, and administrative overhead.

Finney's RPoW system differed from a PoW system in permitting the random exchange of tokens without repeating the work required to generate them. After someone had "spent" a PoW token at a website, the website's operator could exchange that "spent" PoW token for a new, unspent RPoW token, which could then be spent at some third-party website similarly equipped to accept RPoW tokens. This would save the resources otherwise needed to 'mint' a PoW token. The anti-counterfeit property of the RPoW token was guaranteed by remote attestation. The RPoW server that exchanges a used PoW or RPoW token for a new one of equal value uses remote attestation to allow any interested party to verify what software is running on the RPoW server. Since the source code for Finney's RPoW software was published (under a BSD-like license), any sufficiently knowledgeable programmer could, by inspecting the code, verify that the software (and, by extension, the RPoW server) never issued a new token except in exchange for a spent token of equal value.

Until 2009, Finney's system was the only RPoW system to have been implemented; it never saw economically significant use.

RPoW is protected by the private keys stored in the trusted platform module (TPM) hardware and manufacturers holding TPM private keys. Stealing a TPM manufacturer's key or obtaining the key by examining the TPM chip itself would subvert that assurance.

In 2009, the Bitcoin network went online. Bitcoin is a proof-of-work cryptocurrency that, like Finney's RPoW, is also based on the Hashcash PoW. But in Bitcoin double-spend protection is provided by a decentralized P2P protocol for tracking transfers of coins, rather than the hardware trusted computing function used by RPoW. Bitcoin has better trustworthiness because it is protected by computation. Bitcoins are "mined" using the Hashcash proof-of-work function by individual miners and verified by the decentralized nodes in the P2P bitcoin network.

The difficulty is periodically adjusted to keep the block time around a target time.

Within the Bitcoin community there are groups working together in mining pools.^[19] Some miners use ASICs (Application-Specific Integrated Circuit) for PoW.^[20] Some PoWs claim to be ASIC resistant, i.e. to limit the efficiency gain that an ASIC can have over commodity hardware, like a GPU, to be well under an order of magnitude. In practice, such claims have not withstood the test of time.