0
$\begingroup$

Let $U$ and $P$ be prefix-free languages with alphabet $\{0,1\}$. We say that $U$ can encode $P$ with at most a constant overhead if there exists an injective function $c:P \to U$ and a constant $a$ such that

$$\forall s \in P. |c(s)| \le a + |s|$$

For example, any prefix-free language can encode itself with an overhead of $0$.

Consider the prefix-free languages

$$U = (00|01)^*1$$ $$P = 0^*1$$

Then let the function $c$ which takes the string $0^n1$, converts $n$ to binary, puts a $0$ before each bit, and puts a $1$ at the end. Then $U$ encodes $P$ using $c$ with an overhead of $2$.

Additionally, given a countable set of prefix languages $S$, we can define

$$U = 0^n1S_n$$

where $S_n$ is the nth prefix language in $S$. Then $U$ can encode $S_n$ with at most an overhead of $n+1$. So $U$ can encode any prefix-free language in $S_n$ with at most constant overhead.

Is there a language $U$ that can encode every $P$ with at most a constant overhead? The constant and encoding function can be different for each $P$, but must exist for each $P$.

Note that the different encoding functions are permitted to overlap. For example, if the encoding functions for $P$ and $P'$ are $c$ and $c'$, then

$$\exists s \in P, s' \in P'. c(s) = c'(s')$$

is permitted. Indeed, it is necessary for many $P$ and $P'$ since there are uncountably many non-empty prefix-free languages but only countably many strings in $U$.

$$\exists s, s' \in P. s \neq s' \land c(s) = c(s')$$

is not permitted.

Additionally, $U$ does not need to be computable, but neither does $P$.

$\endgroup$
0
$\begingroup$

No

Let us say that $U$ is a prefix-free language that can encode any other prefix-free language with at most constant overhead.

Let

$$L_n = \{s \in U: |s| = n\}$$ $$L_{\le n} = \{s \in U: |s| \le n\}$$

for any language $L$.

By Kraft's inequality $$\sum_{i=0}^\infty \frac {|U_i|} {2^i} \le 1$$

This implies for any $\epsilon > 0$, there exists a $n$ such that for every $N > n$ we have $\frac {|U_N|} {2^N} < \frac \epsilon 2$. That implies

$$|U_{\le N}| = |U_{\le n}| + \sum_{i=n+1}^N \frac {|U_i|} {2^i} 2^i < |U_{\le n}| + \sum_{i=n+1}^N \frac \epsilon 2 2^i $$$$ = |U_{\le n}| + \frac \epsilon 2(2^{N+1} - 2^{n+1}) < |U_{\le n}| + \epsilon 2^N$$

which implies $$\forall \epsilon > 0. \lim_{n \to \infty} \frac {|U_{\le n}|} {2^n} \le \epsilon$$ $$\lim_{n \to \infty} \frac {|U_{\le n}|} {2^n} = 0$$

Now, let $f$ be a function such that

$$\frac {|U_{\le 2n + f(n)}|} {2^{2n + f(n)}} \lt 2^{-2n}$$

For any $n$, a $f(n)$ will exist with this property.

Now let $P$ be the prefix-free language

$$P = \{0^{n-1}1t : n > 0, t \in (0|1)^{f(n)}\}$$

There will exist some injective function $c:U \to P$ and constant $a$ such that

$$\forall s \in P. |c(s)| \le a + |s|$$

Also note that $|P_{\le n + f(n)}| \ge 2^{f(n)}$, since there are $2^{f(n)}$ strings with length $f(n)$.

Now

$$|U_{\le 2a + f(a)}| < 2^{2a + f(a) - 2a} = 2^{f(a)} \le |P_{\le a + f(a)}|$$

On the other hand, since $c : P \to U$ is injective

$$\forall n. |P_{\le n}| \le |U_{\le a + n}|$$

A contradiction!

A similar argument can be used to show that there is no computable prefix-free language that encodes any other computable prefix-free language with at most constant overhead. That is because when $U$ is computable, $f$ is computable via a search.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.