Binary Lambda Calculus and Combinatory Logic

John Tromp
May 14, 2014

Abstract
In the first part, we introduce binary representations of both lambda
calculus and combinatory logic terms, and demonstrate their simplicity
by providing very compact parser-interpreters for these binary languages.
Along the way we also present new results on list representations, bracket
abstraction, and fixpoint combinators. In the second part we review Al-
gorithmic Information Theory, for which these interpreters provide a con-
venient vehicle. We demonstrate this with several concrete upper bounds
on program-size complexity.

1 Introduction
The ability to represent programs as data and to map such data back to pro-
grams (known as reification and reflection [10]), is of both practical use in
metaprogramming [15] as well as theoretical use in computability and logic [18].
It comes as no surprise that the pure lambda calculus, which represents both
programs and data as functions, is well equipped to offer these features. In [7],
Kleene was the first to propose an encoding of lambda terms, mapping them
to Gödel numbers, which can in turn be represented as so called Church nu-
merals. Decoding such numbers is somewhat cumbersome, and not particularly
efficient. In search of simpler constructions, various alternative encodings have
been proposed using higher-order abstract syntax [9] combined with the stan-
dard lambda representation of signatures [12]. A particularly simple encoding
was proposed by Mogensen [24], for which the term λm.m(λx.x)(λx.x) acts as
a selfinterpreter. The prevalent data format, both in information theory and
in practice, however, is not numbers, or syntax trees, but bits. We propose
binary encodings of both lambda and combinatory logic terms, and exhibit rel-
atively simple and efficient interpreters (using the standard representation of
bit-streams as lists of booleans).
This gives us a representation-neutral notion of the size of a term, measured
in bits. More importantly, it provides a way to describe arbitrary data with, in a
sense, the least number of bits possible. We review the notion of how a computer
reading bits and outputting some result constitutes a description method, and
how universal computer correspond to optimal description methods. We then

1
pick specific universal computers based on our interpreters and prove several of
the basic results of Algorithmic Information Theory with explicit constants.

2 Lambda Calculus
We only summarize the basics here. For a comprehensive treatment we refer
the reader to the standard reference [19].
Assume a countably infinite set of variables

a, b, . . . , x, y, z, x0 , x1 , . . .

The set of lambda terms Λ is built up from variables using abstraction

(λx.M )

and application
(M N ),
where x is any variable and M, N are lambda terms. (λx.M ) is the function
that maps x to M , while (M N ) is the application of function M to argument
N . We sometimes omit parentheses, understanding abstraction to associate to
the right, and application to associate to the left, e.g. λx.λy.x y x denotes
(λx.(λy.((x y)x))). We also join consecutive abstractions as in λx y.x y x.
The free variables F V (M ) of a term M are those variables not bound by
an enclosing abstraction. Λ0 denotes the set of closed terms, i.e. with no free
variables. The simplest closed term is the identity λx.x.
We consider two terms identical if they only differ in the names of bound
variables, and denote this with ≡, e.g. λy.y x ≡ λz.z x. The essence of λ
calculus is embodied in the β-conversion rule which equates

(λx.M )N = M [x := N ],

where M [x := N ] denotes the result of substituting N for all free occurrences

of x in M (taking care to avoid variable capture by renaming bound variables
in M if necessary). For example,

(λx y.y x)y ≡ (λx.(λz.z x))y ≡ (λx z.z x)y = λz.z y.

A term with no β-redex, that is, no subterm of the form (λx.M )N , is said to
be in normal form. Terms may be viewed as denoting computations of which
β-reductions form the steps, and which may halt with a normal form as the end
result.

2.1 Some useful lambda terms

Define (for any M, P, Q, . . . , R)

I ≡ λx.x

2
 true ≡ λx y.x
nil ≡ false ≡ λx y.y
hP, Q, . . . , Ri ≡ λz.z P Q . . . R
M [0] ≡ M true
M [i + 1] ≡ (M false)[i]
Y ≡ λf.((λx.x x)(λx.f (x x)))
Ω ≡ (λx.x x)(λx.x x)
Note that
true P Q = (λx y.x) P Q = x[x := P ] = P
false P Q = (λx y.y) P Q = y[y := Q] = Q,
justifying the use of these terms as representing the booleans.
A pair of terms like P and Q is represented by hP, Qi, which allows one to
retrieve its parts by applying htruei or hfalsei:
htrueihP, Qi = hP, Qi true = true P Q = P
hfalseihP, Qi = hP, Qi false = false P Q = Q.
Repeated pairing is the standard way of representing a sequence of terms:
hP, hQ, hR, . . .iii.
A sequence is thus represented by pairing its first element with its tail—the
sequence of remaining elements. The i’th element of a sequence M may be
selected as M [i]. To wit:
hP, Qi[0] = true P Q = P,
hP, Qi[i + 1] ≡ (hP, Qi false)[i] = Q[i].
The empty sequence, for lack of a first element, cannot be represented by
any pairing, and is instead represented by nil. A finite sequence P, Q, . . . , R
can thus be represented as hP, hQ, h. . . , hR, nili . . .iii.
Our choice of nil allows for the processing of a possible empty list s with
the expression
s M N,
which for s ≡ nil reduces to N , and for s ≡ hP, Qi reduces to M P Q N . In
contrast, Barendregt [14] chose I to represent the empty list, which requires
a more complicated list processing expression like like s (λa b c.c a b) M XN ,
which for s = nil reduces to N M X, and for s ≡ hP, Qi reduces to M P Q X N .
Y is the fixpoint operator, that satisfies
Yf = (λx.f (x x))(λx.f (x x)) = f (Y f ).
This allows one to transform a recursive definition f = . . . f . . . into f =
Y(λf.(. . . f . . .)), which behaves exactly as desired.
Ω is the prime example of a term with no normal form, the equivalence of
an infinite loop.

3
2.2 Binary strings
Binary strings are naturally represented by boolean sequences, where true rep-
resents 0 and false represents 1.

Definition 1 For a binary string s and lambda term M , (s : M ) denotes the

list of booleans corresponding to s, terminated with M . Thus, (s : nil) is the
standard representation of string s.
For example, (011 : nil) ≡ htrue, hfalse, hfalse, niliii represents the string 011.
We represent an unterminated string, such as part of an input stream, as an
open term (s : z), where the free variable z represents the remainder of input.

2.3 de Bruijn notation

In [13], de Bruijn proposed an alternative notation for closed lambda terms using
natural numbers rather than variable names. Abstraction is simply written λM
while the variable bound by the n’th enclosing λ is written as the index n. In
this notation, λx y z.z x y ≡ λ λ λ 0 2 1. It thus provides a canonical notation
for all identical terms. Beta-conversion in this notation avoids variable capture,
but instead requires shifting the indices, i.e. adjusting them to account for
changes in the lambda nesting structure. Since variable/index exchanges don’t
affect each other, it’s possible to mix both forms of notation, as we’ll do later.

2.4 Binary Lambda Calculus

Definition 2 The code for a term in de Bruijn notation is defined inductively
as follows:

b ≡ 1n+1 0
n
d ≡ 00M
λM c
M
d N ≡ 01M
cN b

We call |M
c| the size of M .

For example bI ≡ 0010, false

[ ≡ 000010, t[ rue ≡ 0000110 and λx.x
\x ≡
\ ≡ 00000010, of sizes 4,6,7,8 and 8 bits respectively, are the
00011010, λx.false
5 smallest closed terms.
The main result of this paper is the following
Theorem 1 There is a self-interpreter E of size 210 (which happens to be the
product of the smallest four primes), such that for every closed term M and
terms C, N we have
E C (Mc : N ) = C (λz.M ) N

4
The interpreter works in continuation passing style [16]. Given a continuation
and a bitstream containing an encoded term, it returns the continuation applied
to the abstracted decoded term and the remainder of the stream. The reason
for the abstraction becomes evident in the proof.
The theorem is a special case of a stronger one that applies to arbitrary de
Bruijn terms. Consider a de Bruijn term M in which an index n occurs at a
depth of i ≤ n nested lambda’s. E.g., in M ≡ λ3, the index 3 occurs at depth 1.
This index is like a free variable in that it is not bound within M . The interpreter
(being a closed term) applied to other closed terms, cannot produce anything
but a closed term. So it cannot possibly reproduce M . Instead, it produces
terms that expect a list of bindings for free indices. These take the form M z[] ,
which is defined as the result of replacing every free index in M , say n at depth
i ≤ n, by z[n − i]. For example, (λ3)z[] = λz[3 − 1] = λ(z false false true),
selecting binding number 2 from binding list z.
The following claim (using mixed notation) will be needed later.
Claim 1 For any de Bruijn term M , we have (λM )z[] = λy.M hy,zi[]
Proof: A free index n at depth i ≤ n in M , gets replaced by hy, zi[n − i] on
the right. If i < n then n is also free in λM at depth i + 1 and gets replaced
by z[n − i − 1] = hy, zi[n − i]. If i = n then n is bound by the front λ, while
hy, zi[n − i] = hy, zi[0] = y.
To prove Theorem 1 it suffices to prove the more general:
Theorem 2 There is a self-interpreter E of size 210, such that for all terms
M, C, N we have
E C (M c : N ) = C (λz.M z[] ) N

Proof: We take

E ≡ Y (λe c s.s (λa t.t (λb.a E0 E1 )))

E0 ≡ e (λx.b (c (λz y.x hy, zi))(e (λy.c (λz.x z (y z)))))
E1 ≡ (b (c (λz.z b))(λs.e (λx.c (λz.x (z b))) t))

of size 217 and note that the beta reduction from Y M to (λx.x x)(λx.M (x x))
saves 7 bits.
Recall from the discussion of Y that the above is a transformed recursive
definition where e will take the value of E.
Intuitively, E works as follows. Given a continuation c and sequence s, it
extracts the leading bit a and tail t of s, extracts the next bit b, and selects E0
to deal with a = true (abstraction or application), or E1 to deal with a = false
(an index).
E0 calls E recursively, extracting a decoded term x. In case b = true
(abstraction), it prepends a new variable y to bindings list z, and returns the
continuation applied to the decoded term provided with the new bindings. In
case b = false (application), it calls E recursively again, extracting another

5
decoded term y, and returns the continuation applied to the application of the
decoded terms provided with shared bindings.
E1 , in case b = true, decodes to the 0 binding selector. In case b = false,
it calls E recursively on t (coding for an index one less) to extract a binding
selector x, which is provided with the tail z b of the binding list to obtain the
correct selector.
We continue with the formal proof, using induction on M .
Consider first the case where M = 0. Then

E C (M
c : N) = E C (10 : N )
= hfalse, htrue, N ii (λa t.t (λb.a E0 E1 ))
= htrue, N i (λb.false E0 E1 )
= (E1 N )[b := true]
= C (λz.z true) N,

as required. Next consider the case where M = n + 1. Then, by induction,

E C (M
c : N) = E C (1n+2 0 : N )
= hfalse, hfalse, (1n 0 : N )ii (λa t.t (λb.a E0 E1 ))
= (λs.e (λx.C (λz.x (z false)))(1n+1 0 : N ))(1n 0 : N )
= E (λx.C (λz.x (z false))) (b
n : N)
= (λx.C (λz.x (z false))) (λz.nz[] ) N
= C (λz.n(z false)[] ) N
= C (λz.(z false)[n])) N
= C (λz.z[n + 1])) N
= C (λz.(n + 1)z[] ) N,

as required. Next consider the case M = λM 0 . Then, by induction and claim 1,

d0 : N ))
E C ((λM = E C (00M c0 : N )
c0 : N )ii (λa t.t (λb.a E0 E1 ))
= htrue, htrue, (M
= e (λx.(C (λz y.xhy, zi))) (Mc0 : N )
= (λx.(C (λz y.x hy, zi)))(λz.M 0z[] ) N
= C (λz y.(λz.M 0z[] ) hy, zi) N
= C (λz.(λy.M 0hy,zi[] )) N
= C (λz.(λM 0 )z[] )) N,

as required. Finally consider the case M = M 0 M 00 . Then, by induction,

0 M 00 : N )
E C (M\ = E C (01M c0 Md00 : N )
= c0 M
htrue, hfalse, (M d00 : N )ii(λa t.t (λb.a E0 E1 ))
= e (λx.(e (λy.C (λz.x z (y z))))) (M c0 M d00 : N )

6
 = (λx.(e (λy.C (λz.x z (y z)))))(λz.M 0z[] ) (Md00 : N )
= e (λy.C (λz.(λz.M 0z[] ) z (y z)))(M
d00 : N )
= (λy.C (λz.M 0z[] (y z)))(λz.M 00z[] ) N
= C (λz.M 0z[] M 00z[] ) N
= C (λz.(M 0 M 00 )z[] ) N,

as required. This completes the proof of Theorem 1.

In private communication, Bertram Felgenhauer pointed out that replacing
E by Y (λe c s.s (λa t e c.t (λb.a E0 E1 ))ec) and doing one more beta reduction
of (λx.M (x x)), saves another 4 bits to get 206. We conjecture that any self-
interpreter for any binary representation of lambda calculus must be at least 24
bytes in size, which would make E close to optimal.

3 Combinatory Logic
Combinatory Logic (CL) is the equational theory of combinators—terms built
up, using application only, from the two constants K and S, which satisfy

S M N L = M L (N L)
KM N = M

CL may be viewed as a subset of lambda calculus, in which K ≡ λx y.x,

S ≡ λx y z.x z (y z), and where the beta conversion rule can only be applied
groupwise, either for an S with 3 arguments, or for a K with 2 arguments. Still,
the theories are largely the same, becoming equivalent in the presence of the
rule of extensionality (which says M = M 0 if M N = M 0 N for all terms N ).
A process known as bracket abstraction allows for the translation of any
lambda term to a combination—a CL term containing variables in addition to
K and S. It is based on the following identities, which are easily verified:

λx.x = I = SKK
λx.M = KM (x not free in M )
λx.M N = S (λx.M ) (λx.N )

λ’s can thus be successively eliminated, e.g.:

λx y.y x ≡ λx (λy.y x)
= λx (S I(K x))
= S (K (S I))(S (K K) I),

where I is considered a shorthand for S K K.

Bracket abstraction is an operation λ0 on combinations M with respect to a
variable x, such that the resulting combination contains no occurrence of x and

7
behaves as λx.M :

λ0 x. x ≡ I
λ0 x. M ≡ KM (x 6∈ M )
λ x. (M N ) ≡ S (λ x. M ) (λ0 x. N )
0 0

3.1 Binary Combinatory Logic

Combinators have a wonderfully simple encoding as binary strings: encode S as
00, K as 01, and application as 1.

Definition 3 We define the encoding C

e of a combinator C as

K
e ≡ 00
S
e ≡ 01
CgD ≡ 1C
eDe

Again we call |C|

e the size of combinator C.

For instance, the combinator S(KSS) ≡ (S((KS)S)) is encoded as 10111000101.

The size of a combinator with n K/S’s, which necessarily has n−1 applications,
is thus 2n + n − 1 = 3n − 1.
For such a simple language we expect a similarly simple interpreter.
Theorem 3 There is a cross-interpreter F of size 119, such that for every
combinator M and terms C, N we have

F C (M
f : N) = C M N

Proof: We take

F ≡ Y (λe c s.s(λa.a F0 F1 ))
F0 ≡ λt.t (λb.c (b K S))
F1 ≡ e (λx.e (λy.(c (x y))))

of size 131 and note that a toplevel beta reduction saves 7 bits in size, while
replacing K by b saves another 5 bits (we don’t define F that way because of
its negative impact on bracket abstraction).
Given a continuation c and sequence s, it extracts the leading bit a of s, and
tail t extracts the next bit b, and selects F0 to deal with a = true (K or S), or
F1 to deal with a = false (application). Verification is straightforward and left
as an exercise to the reader.
We conjecture that any self-interpreter for any binary representation of com-
binatory logic must be at least 14 bytes in size. The next section considers
translations of F which yield a self-interpreter of CL.

8
3.2 Improved bracket abstraction
The basic form of bracket abstraction is not particularly efficient. Applied to
F, it produces a combinator of size 536.
A better version is λ1 , which uses the additional rule

λ1 x. (M x) ≡ M (x 6∈ M )

whenever possible. Now the size of F as a combinator is only 281, just over half
as big.
Turner [25] noticed that repeated use of bracket abstraction can lead to a
quadratic expansion on terms such as

X ≡ λa b . . . z.(a b . . . z) (a b . . . z),

and proposed new combinators to avoid such behaviour. We propose to achieve a

similar effect with the following set of 9 rules in decreasing order of applicability:

λ2 x. (S K M ) ≡ S K (for all M )
λ2 x. M ≡ KM (x 6∈ M )
2
λ x. x ≡ I
2
λ x. (M x) ≡ M (x 6∈ M )
λ2 x. (x M x) ≡ λ2 x. (S S K x M )
λ2 x. (M (N L)) ≡ λ2 x. (S (λ2 x. M ) N L) (M, N combinators)
λ2 x. ((M N ) L) ≡ λ2 x. (S M (λ2 x. L) N ) (M, L combinators)
λ2 x. ((M L) (N L)) ≡ λ2 x. (S M N L) (M, N combinators)
λ2 x. (M N ) ≡ S (λ2 x. M ) (λ2 x. N )

The first rule exploits the fact that S K M behaves as identity, whether M
equals K, x or anything else. The fifth rule avoids introduction of two Is. The
sixth rule prevents occurrences of x in L from becoming too deeply nested,
while the seventh does the same for occurrences of x in N . The eighth rule
abstracts an entire expression L to avoid duplication. The operation λ2 x. M
for combinators M will normally evaluate to K M , but takes advantage of
the first rule by considering any S K M a combinator. Where λ1 gives an X
combinator of size 2030, λ2 brings this down to 374 bits.
For F the improvement is more modest, to 275 bits. For further improve-
ments we turn our attention to the unavoidable fixpoint operator.
Y, due to Curry, is of minimal size in the λ calculus. At 25 bits, it’s 5 bits
shorter than Turing’s alternative fixpoint operator

Y0 ≡ (λz.z z)(λz.λf.f (z z f )).

But these translate to combinators of size 65 and 59 bits respectively.

In comparison, the fixpoint operator

Y00 ≡ (λx y.x y x)(λy x.y(x y x))

9
translates to combinator

S S K (S (K (S S (S (S S K)))) K)

of size 35, the smallest possible fixpoint combinator as verified by exhaustive

search by computer.
(The situation is similar for Ω which yields a combinator of size 41, while
S S K (S (S S K)), of size 20, is the smallest unsolvable combinator—the
equivalent of an undefined result, see [19]).
Using Y00 instead of Y gives us the following
Theorem 4 There is a self-interpreter F for Combinatory Logic of size 263.
Comparing theorems 3 and 4, we conclude that λ-calculus is a much more
concise language than CL. Whereas in binary λ-calculus, an abstraction takes
only 2 bits plus i + 1 bits for every occurrence of the variable at depth i, in
binary CL the corresponding bracket abstraction typically introduces at least
one, and often several S’s and K’s (2 bits each) per level of depth per variable
occurrence.

4 Program Size Complexity

Intuitively, the amount of information in an object is the size of the shortest
program that outputs the object. The first billion digits of π for example,
contain little information, since they can be calculated by a program of a few
lines only. Although information content may seem to be highly dependent
on choice of programming language, the notion is actually invariant up to an
additive constant.
The theory of program size complexity, which has become known as Algo-
rithmic Information Theory or Kolmogorov complexity after one of its founding
fathers, has found fruitful application in many fields such as combinatorics,
algorithm analysis, machine learning, machine models, and logic.
In this section we propose a concrete definition of Kolmogorov complexity
that is (arguably) as simple as possible, by turning the above interpreters into
a ‘universal computer’.
Intuitively, a computer is any device that can read bits from an input stream,
perform computations, and (possibly) output a result. Thus, a computer is a
method of description in the sense that the string of bits read from the input
describes the result. A universal computer is one that can emulate the behaviour
of any other computer when provided with its description. Our objective is to
define, concretely, for any object x, a measure of complexity of description
C(x) that shall be the length of its shortest description. This requires fixing
a description method, i.e. a computer. By choosing a universal computer, we
achieve invariance: the complexity of objects is at most a constant greater than
under any other description method.
Various types of computers have been considered in the past as description
methods.

10
 Turing machines are an obvious choice, but turn out to be less than ideal:
The operating logic of a Turing machine—its finite control—is of an irregular
nature, having no straightforward encoding into a bitstring. This makes con-
struction of a universal Turing machine that has to parse and interpret a finite
control description quite challenging. Roger Penrose takes up this challenge in
his book [1], at the end of Chapter 2, resulting in a universal Turing machine
whose own encoding is an impressive 5495 bits in size, over 26 times that of E.
The ominously named language ‘Brainfuck’ which advertises itself as “An
Eight-Instruction Turing-Complete Programming Language” [23], can be con-
sidered a streamlined form of Turing machine. Indeed, Oleg Mazonka and Daniel
B. Cristofani [17] managed to write a very clever BF self-interpreter of only 423
instructions, which translates to 423 · log(8) = 1269 bits (the alphabet used is
actually ASCII at 7 or 8 bits per symbol, but the interpreter could be redesigned
to use 3-bit symbols and an alternative program delimiter).
In [5], Levin stresses the importance of a (descriptional complexity) measure,
which, when compared with other natural measures, yields small constants, of
at most a few hundred bits. His approach is based on constructive objects
(c.o.’s) which are functions from and to lower ranked c.o.’s. Levin stops short of
exhibiting a specific universal computer though, and the abstract, almost topo-
logical, nature of algorithms in the model complicates a study of the constants
achievable.
In [2], Gregory Chaitin paraphrases John McCarthy about his invention of
LISP, as “This is a better universal Turing machine. Let’s do recursive function
theory that way!” Later, Chaitin continues with “So I’ve done that using LISP
because LISP is simple enough, LISP is in the intersection between theoretical
and practical programming. Lambda calculus is even simpler and more elegant
than LISP, but it’s unusable. Pure lambda calculus with combinators S and K,
it’s beautifully elegant, but you can’t really run programs that way, they’re too
slow.”
There is however nothing intrinsic to λ calculus or CL that is slow; only
such choices as Church numerals for arithmetic can be said to be slow, but
one is free to do arithmetic in binary rather than in unary. Frandsen and
Sturtivant [11] amply demonstrate the efficiency of λ calculus with a linear
time implementation of k-tree Turing Machines. Clear semantics should be a
primary concern, and Lisp is somewhat lacking in this regard [4]. This paper
thus develops the approach suggested but discarded by Chaitin.

4.1 Functional Complexity

By providing the appropriate continuations to the interpreters that we con-
structed, they become universal computers describing functional terms modulo
equality. Indeed, for

U ≡ E hΩi
0
U ≡ FI

11
of sizes |U| f0 | = 272, Theorems 1 and 3 give
b = 236 and |U

U (Mc : N) = M N
U0 (M
f : N) = M N

for every closed λ-term or combinator M and arbitrary N , immediately estab-

lishing their universality.
The universal computers essentially define new binary languages, which
we may call universal binary lambda calculus and universal combinatory logic,
whose programs comprise two parts. The first part is a program in one of the
original binary languages, while the second part is all the binary data that is
consumed when the first part is interpreted. It is precisely this ability to embed
arbitrary binary data in a program that allows for universality.
Note that by Theorem 2, the continuation hΩi in U results in a term M Ω[] .
For closed M , this term is identical to M , but in case M is not closed, a free
index n at λ-depth n is now bound to Ω[n − n], meaning that any attempt
to apply free indices diverges. Thus the universal computer essentially forces
programs to be closed terms.
We can now define the Kolmogorov complexity of a term x, which comes in
three flavors. In the simple version, programs are terminated with N = nil and
the result must equal x. In the prefix version, programs are not terminated,
and the result must equal the pair of x and the remainder of the input. In both
cases the complexity is conditional on zero or more terms yi .

Definition 4

KS(x|y1 , . . . , yk ) = min{l(p) | U (p : nil) y1 . . . yk = x}

KP (x|y1 , . . . , yk ) = min{l(p) | U (p : z) y1 . . . yk = hx, zi}

The definition also applies to infinite terms according to the infinitary lambda
calculus of [20].
In the special case of k = 0 we obtain the unconditional complexities KS(x)
and KP (x).
Finally, for a binary string s, we can define its monotone complexity as

KM (s|y1 , . . . , yk ) = min{l(p) | ∃M : U (p : Ω) y1 . . . yk = (s : M )}.

In this version, we consider the partial outputs produced by increasingly longer

prefixes of the input, and the complexity of s is the shortest program that causes
the output to have prefix s.

4.2 Monadic IO
The reason for preserving the remainder of input in the prefix casse is to facilitate
the processing of concatenated descriptions, in the style of monadic IO [21].

12
Although a pure functional language like λ calculus cannot define functions with
side effects, as traditionally used to implement IO, it can express an abstract
data type representing IO actions; the IO monad. In general, a monad consists
of a type constructor and two functions, return and bind (also written >>= in
infix notation) which need to satisfy certain axioms [21]. IO actions can be seen
as functions operating on the whole state of the world, and returning a new state
of the world. Type restrictions ensure that IO actions can be combined only
through the bind function, which according to the axioms, enforces a sequential
composition in which the world is single-threaded. Thus, the state of the world
is never duplicated or lost. In our case, the world of the universal machine
consists of only the input stream. The only IO primitive needed is readBit,
which maps the world onto a pair of the bit read and the new world. But a list
is exactly that; a pair of the first element and the remainder. So readBit is
simply the identity function! The return function, applied to some x, should
map the world onto the pair of x and the unchanged world, so it is defined
by return ≡ λx y.hx, yi. Finally, the bind function, given an action x and
a function f , should subject the world y to action x (producing some ha, y 0 i)
followed by action f a, which is defined by bind ≡ λx f y.x y f (note that
ha, y 0 if = f a y 0 ) One may readily verify that these definitions satisfy the
monad axioms. Thus, we can wite programs for U either by processing the
input stream explicitly, or by writing the program in monadic style. The latter
can be done in the pure functional language ‘Haskell’ [22], which is essentially
typed lambda calculus with a lot of syntactic sugar.

4.3 An Invariance Theorem

The following theorem is the first concrete instance of the Invariance Theorem,
2.1.1 in [6].

Theorem 5 Define KS 0 (x|y1 , . . . , yk ) and KP 0 (x|y1 , . . . , yk ) analogous to Def-

inition 4 in terms of U0 . Then KS(x) ≤ KS 0 (x) + 125 and KP (x) ≤ KP 0 (x) +
125.

The proof is immediate from Theorem 3 by using U c0 of length 125 as prefix

to any program for U0 . We state without proof that a redesigned U translates
to a combinator of size 617, which thus forms an upper bound in the other
direction.
Now that complexity is defined for as rich a class of objects as terms (modulo
equality), it is easy to extend it to other classes of objects by mapping them
into λ terms.
For binary strings, this means mapping string s onto the term (s : nil). And
for a tuple of binary strings s0 , . . . , sk , we take h(s0 : nil), . . . , (sk : nil)i.
We next look at numbers in more detail, revealing a link with self-delimiting
strings.

13
4.4 Numbers and Strings
Consider the following correspondence between natural numbers, binary strings,
and strings over {1, 2}:

n∈N: 0 1 2 3 4 5 6 7 8 9 ...
x ∈ {0, 1}∗ :  0 1 00 01 10 11 000 001 010 ...
y ∈ {1, 2}∗ :  1 2 11 12 21 22 111 112 121 ...

in which the number n corresponds to

• the n-th binary string xin lexicographic order
• the string x obtained by stripping the leading 1 from (n + 1)2 , the binary
representation of n + 1
• the string y = n{1,2} , the base 2 positional system using digits {1, 2}.
Pl−1 Pl−1 Pl−1 Pl−1
Indeed, n + 1 = 2l + i=0 xi 2i iff n = i=0 2i + i=0 xi 2i iff n = i=0 (xi +
1)2i .

4.5 Prefix codes

Another way to tie the natural numbers and binary strings together is by what
we shall call the binary natural tree shown in Figure 1. It has the set of natural
numbers as vertices, and the set of binary strings as edges, such that the 2n
length-n strings are the edges leading from vertex n. Edge w leads from vertex
|w| to w + 1, which in binary is 1w.
In [8], Vladimir Levenshtein defines a universal code for the natural numbers
that corresponds to concatenating the edges on the path from 0 to n, prefixed
with a unary encoding of the depth of vertex n in the tree. The resulting set of
codewords is prefix-free. meaning that no string is a proper prefix of another,
which is the same as saying that the strings
P in the set are self-delimiting. Prefix-
−|s|
free sets satisfy the Kraft inequality: s2 ≤ 1. We’ve already seen two
important examples of prefix-free sets, namely the set of λ term encodings M c
and the set of combinator encodings M . The Levenshtein code n of a number
f
n can be defined recursively as

0=0 n + 1 = 1 l(n) n.

and satisfies the following nice properties:

• prefix-free and complete: n≥0 2−|n| = 1.

P

• identifies lexicographic with numeric order: m lexicographically precedes

n if and only if m < n.
• simple to encode and decode

14
 Figure 1: binary natural tree

Figure 2: codes on the unit interval; 0 = 0, 1 = 10, 2 = 110 0, 3 = 110 1, 4 =

1110 0 00, 5 = 1110 0 01, 6 = 1110 0 10, 7 = 1110 0 11, 8 = 1110 1 000, etc..

• efficient in that for every k: |n| ≤ l(n) + l(l(n)) + · · · + lk−1 (n) + O(lk (n)),
where l(s) denotes the length of a string s.

Figure 2 shows the codes as segments of the unit interval, where code x covers
all the real numbers whose binary expansion starts as 0.x, and lexicographic
order translates into left-to-right order.

5 Upper bounds on complexity

Having provided concrete definitions of all key ingredients of algorithmic infor-
mation theory, it is time to prove some concrete results about the complexity
of strings.
The simple complexity of a string is upper bounded by its length:

KS(x) ≤ |bI| + l(x) = l(x) + 4

The prefix complexity of a string is upper bounded by the length of its

delimited version:

KP (x) ≤ |delimit|
\ + l(x) = l(x) + 338.

where delimit is an optimized translation of the following code

let
id = \x x;
nil = \x\y y;
-- readbit cont church_(n+1) list returns cont church_(n_list+1)
-- where n_list is the number corresponding to the concatenation of n and list
readbit = \cont\n\list list (\bit cont (\f\x (n f (n f (bit x (f x))))));

15
 -- dlmt cont list returns cont church_n rest_list
dlmt = \cont\list list (\bit bit (cont nil) (dlmt (\len len readbit cont id)));
-- incc cont done list return either cont (if carry) or done (if no carry) of incremented list
incc = \cont\done\list list (\msb incc (\r\_ msb done cont (\z z (\x\y msb y x) r))
(\r\_ done (\z z msb r))
) (cont list);
-- inc list returns incremented list
inc = incc (\r\z z (\x\y x) r) id;
in dlmt (\n\l\z z (n inc nil) l)

The prefix complexity of a pair is upper bounded by the sum of individual

prefix complexities, one of which is conditional on the shortest program of the
other:
K(x, y) ≤ K(x) + K(y|x∗ ) + 657.
This is the easy side of the fundamental “Symmetry of information” theorem
K(x) − K(x|y ∗ ) = K(y) − K(y|x∗ ) + O(1), which says that y contains as much
information about x as x does about y.
In [3], Chaitin proves the same theorem using a resource bounded evaluator,
which in his version of LISP comes as a primitive called ”try”. His proof is
embodied in the program gamma:

((’ (lambda (loop) ((’ (lambda (x*) ((’ (lambda (x) ((’ (lambda (y) (cons x
(cons y nil)))) (eval (cons (’ (read-exp)) (cons (cons ’ (cons x* nil))
nil)))))) (car (cdr (try no-time-limit (’ (eval (read-exp))) x*)))))) (loop
nil)))) (’ (lambda (p) (if(= success (car (try no-time-limit (’ (eval
(read-exp))) p))) p (loop (append p (cons (read-bit) nil)))))))

of length 2872 bits.

We constructed an equivalent of ”try” from scratch. The constant 657 is the
size of the term pairup defined below, containing a monadic lambda calculus
interpreter that tracks the input bits read so far (which due to space restrictions
is only sparsely commented):

let
id = \x x;
true = \x \y x;
false = \x \y y;
nil = false;
-- monadic parser
uni = \abs \app let uni0 = \cnt \ps \xs
xs (\b0 let ps0 = \ts ps (\p p b0 ts) in
\ys \uni0 \cnt ys (\b1
let ps1 = \ts ps0 (\p p b1 ts) in
b0 (uni0 (\v1 (b1 (cnt (\ctx abs (\v2 v1 (\p p v2 ctx))))
(uni0 (\v2 cnt (\ctx app (v1 ctx) (v2 ctx)))))))
(b1 (cnt (\ctx ctx b1))
(\d \d uni0 (\v cnt (\ctx v (ctx b1))) ps0 ys)) ps1)) uni0 cnt
in uni0;
-- this monad pairs up lambdas with some attribute

16
 abs = \v \p v; app = \ca ca false;
-- pair each suffix with a prefix attribute
list = \x \xs \ys
let mx = abs (\e abs (x e));
mxs = xs list (\rs ys (\p p x rs)) in
\p p ys (\e e false mx false mxs);
pairup = uni abs app (\prog \ps \pi
let pspi = \d (
prog d -- provide ctx
d -- monadic apply
(pi list ps) -- to monadic input, expect monadic <poM,qsM>
false -- get <poM,qsM>; can’t use d here in case prog is id
(\p \e \p \e e) ) -- monadic tail
true -- head of qsM is difflist of bits read
nil); -- terminate list
-- Direct evaluator, using the identity monad.
eval = uni id id id pi
in eval (ps pi) (\po \qs eval qs (pspi qs) (\qo \rs \p p (\p p po qo) rs))) id;
in pairup

Although more involved, our program is less than a quarter the size of
Chaitin’s when measured in bits. Chaitin also offered a program of size 2104
bits, at the cost of introducing yet another primitive into his language, which is
still 220% longer than ours.

6 Future Research
It would be nice to have an objective measure of the simplicity and expres-
siveness of a universal machine. Sizes of constants in fundamental theorems
are an indication, but one that is all too easily abused. Perhaps diophantine
equations can serve as a non-arbitrary language into which to express the com-
putations underlying a proposed definition of algorithmic complexity, as Chaitin
has demonstrated for relating the existence of infinitely many solutions to the
random halting probability Ω. Speaking of Ω, our model provides a well-defined
notion of halting as well, namely when U (p : z) = hM, zi for any closed term
M with a normal form. Computing upper and lower bounds on the value of
Ωλ , as Chaitin did for his LISP-based Ω, and Calude et al. for various other
languages, should be of interest as well. A big task remains in finding a good
constant for the other direction of the ‘Symmetry of Information’ theorem, for
which Chaitin has sketched a program. That constant is bigger by an order of
magnitude, making its optimization an everlasting challenge.

7 Conclusion
The λ-calculus is a surprisingly versatile and concise language, in which not
only standard programming constructs like bits, tests, recursion, pairs and lists,

17
but also reflection, reification, and marshalling are readily defined, offering an
elegant concrete foundation of algorithmic information theory.
An implementation of Lambda Calculus, Combinatory Logic, along with
their binary and universal versions, written in Haskell, is available at [26].

8 Acknowledgements
I am greatly indebted to Paul Vitányi for fostering my research into concrete
definitions of Kolmogorov complexity, and to Robert Solovay, Christopher Hen-
drie and Bertram Felgenhauer for illuminating discussions on my definitions and
improvements in program sizes.

References
[1] R. Penrose, The Emperor’s New Mind, Oxford University press, 1989.

[2] G. Chaitin, An Invitation to Algorithmic Information Theory,

DMTCS’96 Proceedings, Springer Verlag, Singapore, 1997, pp. 1–23
(http://www.cs.auckland.ac.nz/CDMTCS/chaitin/inv.html).
[3] G. Chaitin, Exploring Randomness, Springer Verlag, 2001.
(http://www.cs.auckland.ac.nz/CDMTCS/chaitin/ait3.html)
[4] R. Muller, M-LISP: A representation-independent dialect of LISP with re-
duction semantics, ACM Transactions on Programming Languages and Sys-
tems 14(4), 589–616, 1992.
[5] L. Levin, On a Concrete Method of Assigning Complexity Measures, Doklady
Akademii nauk SSSR, vol. 18(3), pp. 727–731, 1977.
[6] M. Li and P. Vitányi, An Introduction to Kolmogorov Complexity and Its
Applications, Graduate Texts in Computer Science, second edition, Springer-
Verlag, New York, 1997.

[7] S.C. Kleene, Lambda-Definability and Recursiveness, Duke Mathematical

Journal, 2, 340–353, 1936.
[8] Vladimir Levenshtein, On the redundancy and delay of decodable coding of
natural numbers, Systems Theory Research, 20, 149–155, 1968.
[9] Frank Pfenning and Conal Elliot, Higher-Order Abstract Syntax, ACM SIG-
PLAN’88 Conference on Programming Language Design and Implementa-
tion, 199–208, 1988.
[10] D. Friedman and M. Wand, Reification: Reflection without Metaphysics,
Proc. ACM Symposium on LISP and Functional Programming, 348–355,
1984.

18
[11] Gudmund S. Frandsen and Carl Sturtivant, What is an Efficient Imple-
mentation of the λ-calculus?, Proc. ACM Conference on Functional Pro-
gramming and Computer Architecture (J. Hughes, ed.), LNCS 523, 289–312,
1991.
[12] J. Steensgaard-Madsen, Typed representation of objects by functions,
TOPLAS 11-1, 67–89, 1989.
[13] N.G. de Bruijn, Lambda calculus notation with nameless dummies, a tool
for automatic formula manipulation, Indagationes Mathematicae 34, 381–
392, 1972.
[14] H.P. Barendregt, Discriminating coded lambda terms, in (A. Anderson and
M. Zeleny eds.) Logic, Meaning and Computation, Kluwer, 275–285, 2001.
[15] Franois-Nicola Demers and Jacques Malenfant, Reflection in logic, func-
tional and object-oriented programming: a Short Comparative Study, Proc.
IJCAI Workshop on Reflection and Metalevel Architectures and their Ap-
plications in AI, 29–38, 1995.
[16] Daniel P. Friedman, Mitchell Wand, and Christopher T. Haynes, Essentials
of Programming Languages – 2nd ed, MIT Press, 2001.
[17] Oleg Mazonka and Daniel B. Cristofani, A Very Short Self-Interpreter,
http://arxiv.org/html/cs.PL/0311032, 2003.
[18] D. Hofstadter, Godel, Escher, Bach: an Eternal Golden Braid, Basic Books,
Inc., 1979.
[19] H.P. Barendregt, The Lambda Calculus, its Syntax and Semantics, revised
edition, North-Holland, Amsterdam, 1984.
[20] J.R. Kennaway and J. W. Klop and M.R. Sleep and F.J. de Vries, Infinitary
Lambda Calculus, Theoretical Compututer Science, 175(1), 93–125,1997.
[21] Simon Peyton Jones, Tackling the awkward squad: monadic input/output,
concurrency, exceptions, and foreign-language calls in Haskell, in ”Engineer-
ing theories of software construction”, ed. Tony Hoare, Manfred Broy, Ralf
Steinbruggen, IOS Press, 47–96, 2001.
[22] The Haskell Home Page, http://haskell.org/.
[23] Brainfuck homepage, http://www.muppetlabs.com/~breadbox/bf/.
[24] Torben Æ. Mogensen, Linear-Time Self-Interpretation of the Pure Lambda
Calculus, Higher-Order and Symbolic Computation 13(3), 217-237, 2000.
[25] D. A. Turner, Another algorithm for bracket abstraction, J. Symbol. Logic
44(2), 267–270, 1979.
[26] J. T. Tromp, http://www.cwi.nl/~tromp/cl/cl.html, 2004.

19