Pi Calculus
The Pi calculus is a process calculus invented by Robin Milner in 1992. It is based on channels which can be used to transmit data, and processes which determine the behavior of those channels. It is similar to lambda calculus in that there is only one first-class datatype, but pi calculus also allows concurrent execution, stateful functions, and a richer structure for programs.
Introduction
In the basic form of the calculus, processes are built recursively from the following commands:
Process | Description | Read as | Meaning |
---|---|---|---|
x(y).P |
Input | Read from the channel x, then execute P with the result named y | |
x<y>.P |
Output | Write y to the channel x, then execute P | |
(v x).P |
Creation | new x in P | Execute P with a new channel named x |
P|Q |
Concurrency | Par | Execute P and Q concurrently |
!P |
Replication | Bang | Execute P an infinite number of times concurrently |
0 |
Nil | Does nothing |
Commands in a process are generally separated with .
, and processes are terminated with a 0
(although this is often omitted to improve readability).
Reads implement pattern-matching, meaning that the rest of the process will not execute if the current value taken from the channel does not satisfy the condition. A name acts like a wildcard, and any current value in the channel will match a name. Pattern-matches towards a literal will only be satisfied if the current value in the channel is exactly the literal specified in the read.
Importantly, both read and writes will block the current process from continuing further until they complete. In this way, communication can be used to block processes from running until particular conditions are met. Replication is often restricted to just before an input operation, as in !x(y)
(termed a "replicated read"); this does not weaken the calculus.
Providers
Unlike lambda calculus, which is entirely based around functions, pi calculus does not have any primitive notion of a function. Despite this, it is possible to make a "provider", which behaves similarly to a function. Providers are processes which constantly read from a particular channel and perform some operation whenever data is transmitted. A provider is typically written like this:
!provider(a).a(input1).a(input2). P .a<output1>.a<output2>.0
To later invoke this provider:
(v a).provider<a>.a<input1>.a<input2>.a(output1).a(output2)
Since the provider can be invoked multiple times concurrently, each invocation uses a separate temporary communication channel a
instead of transmitting the inputs and outputs directly.
Structural Congruences
P|0 = P
- Executing parallel with the inert process is the same as executing without the process.P|Q = Q|P
- Parallel composition is commutative.(P|Q)|R = P|(Q|R)
- Parallel composition is associative.!P = !P|P
- InfiniteP
in parallel is the same as adding another P in parallel.(v a)(v b)P = (v b)(v a)P
- The definition of new channels is associative.(v c)(P|Q) = (v c)P|Q
, ifc
also appears inQ
c<x>.P|c(y).Q = P|Q
, in which ally
inQ
is replaced withx
(v x)0 = 0
Examples
Cat program
This process endlessly copies anything read from i
, and writes it to o
.
!i(x).o<x>.0
Thread Join
This process will only execute S
once P
, Q
, and R
have finished.
(v x). ( P.x<x>.0 | Q.x<x>.0 | R.x<x>.0 | x(y).x(y).x(y).S )
In Rho-calculus, this can be expressed simply as
( P; Q; R ).S
Add one provider
!incr(a,x).a<x+1>
Extensions
The pi calculus is sometimes extended for notational convenience, or to make it easier to express certain concepts.
Polyadic communication
One common extension permits read and write operations to accept more than one argument. An example of a polyadic write operation would be f<x,y,z>
, which can be interpreted as shorthand for (v a).f<a>.a<x>.a<y>.a<z>
. Likewise, a polyadic read operation f(x,y,z)
could be interpreted as f(a).a(x).a(y).a(z)
.
Nondeterministic choice operator
The nondeterministic choice operator P+Q
will execute one of either P
or Q
, but it is unknown which. This could be modeled as (v a)(a(x).P|a(x).Q|a<a>.0)
.
Match / Mismatch
The operator if x=y then P
behaves like P
if x
and y
are the same. [1]
The operator if x≠y then P
behaves like P
if x
and y
are not the same.
In some variants, if cond then P
can also be written as [cond].P
.
It is possible to simulate if/else via if x=y then P + if x≠y then Q
. This can also be written as if x then P else Q
.
If the equality comparison is made between channel names, then it is possible to simulate the comparison without match operators. For example, [x=y].Q + [x=z].R
can be modeled as x<0> | (y(a).Q + z(a).R)
. This works because if a value is sent through the channel x, and the channel y recieves the message, then the channels x and y are the same channels; and vice versa.
Cryptographic primitives
As the pi calculus is very good at modeling communication systems, it has been extended to include primitives which cryptographically encrypt and decrypt data streams [2].
Arithmetic and common data structures
Common data structures is a frequent extension, along with their related operations. This may include, but is not limited to, numbers (and arithmetic), lists(maps, filters), sets, maps, etc.
Rho calculus
Rho calculus is a process calculus invented by Greg Meredith which is based on Pi calculus. One of the most significant differences between the two calculi is that channels can not only recieve values and names, they can recieve processes as well (which, in turn, allow programs to modify itself in a way). Another significant difference is that process replication (!P
) has been replaced with recursion.
In Rho calculus, the syntax *x
means to evaluate the content in the name x
as a process in Rho calculus, while the syntax @x
means to un-evaluate the content in x (does not modify x, just returns the value).
The syntax (P;Q;R).a
means to execute a
only if the processes P, Q, and R
have all finished execution.
In Rho calculus, there is a way to peek the frontmost value of a channel, store the value to a name, without consuming the value from the channel. One of the ways to express it is a{x}
, in which the value x
is being 'peeked' from the channel a
. Since variables can be easily expressed by channels, this operation can be expressed more simply as a
.
Encoding of Lambda Calculus
Lambda calculus can be readily encoded using providers. As lambda calculus is functional in nature while pi calculus is procedural in nature, the lambda term must be used as a channel rather than a process. Let [f]
denote a channel representing the lambda term f
, and make these substitutions repeatedly:
c<[λx.f]>.P
⇒(v l)(!l(a).a(x).a<[f]>|c<l>.P)
[λx.f]<c>.P
⇒(v l)(!l(a).a(x).a<[f]>|l<c>.P)
c<[f(x)]>.P
⇒(v a).[f]<a>.a<[x]>.a(r).c<r>.P
[f(x)]<c>.P
⇒(v a).[f]<a>.a<[x]>.a(r).r<c>.P
c<[x]>.P
⇒c<x>.P
(wherex
is atomic)[x]<c>.P
⇒x<c>.P
(wherex
is atomic)
Some care must be taken to ensure that the temporary variables l
, a
, r
don't clash with any other variables created during the expansion, particularly in [f(x)]
terms.
See also
External resources
- Pi calculus at Wikipedia
- FAQ on Pi-Calculus
- A Rho-calculus resource, which also includes Pi calculus recources
- Rholang, a language based on Rho-calculus (which is in turn based on Pi-calculus)
References
- ↑ Parrow, J. [1].
- ↑ Blanchet, Bruno. Modeling and Verifying Security Protocols with the Applied Pi Calculus and ProVerif.