# The Problem

I've been self-studying Introduction to Analysis of Algorithms by Sedgewick and Flajolet. I'm on the fifth chapter, and struggling with exercise 5.1:

How many bit strings of length N have no 000?

More specifically, the authors want a closed form expression counting the number of bit strings of length N which have no 000 as a substring.

# My Plan of Attack

I believe the authors intend the problem to be solved in these steps:

1. Find an "Analytic Combinatorial" expression for the problem statement
2. Convert the aforementioned expression into a generating function
3. Find a corresponding series for the generating function; in particular, find a closed form for the coefficients

The closed form for the coefficients will give me the answer I seek. I believe I have either:

• found a needlessly complicated analytic combinatorial expression, or
• do not understand generating functions enough to solve the generating function

# My Work Thus Far

## The Analytic Combinatoric Expression

I arrived at this form for the generating function:

$$\mathcal{G} = \varepsilon + \mathcal{Z}_0 + \mathcal{Z}_0 \times \mathcal{Z}_0 + (\mathcal{Z}_1 + \mathcal{Z}_0 \times \mathcal{Z}_1 + \mathcal{Z}_0 \times \mathcal{Z}_0 \times \mathcal{Z}_1) \times \mathcal{G}$$

I've used Sedgewick and Flajolet's notation where $\mathcal{G}$ corresponds to the set in which I'm interested, $\mathcal{Z_x}$ corresponds to the singleton set containing the object $x$, and $\varepsilon$ is a "neutral object of size zero".

I justify this expression as follows. Strings that don't have three consecutive zeros can have three different configurations of their three leading bits:

1. $001$
2. $01X$
3. $1XX$

I've used $X$ to indicate either $0$ or $1$.

Therefore, any shorter string not containing three consecutive zeros can be extended by prepending either $1$, $01$, or $001$.

I also add $0$ and $00$ since I cannot generate them using the previously mentioned scheme.

I confirmed that I can generate all length three strings except $000$ using this analytic combinatoric expression:

1. $101 = 1 \cdot 01 \cdot \varepsilon$
2. $100 = 1 \cdot 00$
3. $011 = 01 \cdot 1 \cdot \varepsilon$
4. $010 = 01 \cdot 0$
5. $111 = 1 \cdot 1 \cdot 1 \cdot \varepsilon$
6. $110 = 1 \cdot 1 \cdot 0$
7. $001 = 001 \cdot \varepsilon$

## The Generating Function

From the analytic combinatoric expression I derived this generating function:

$$G(z) = \frac{1 + z + z^2}{1 - z - z^2 - z^3}$$

And this is where I've become quite stuck. AFAIK, the denominator, $1 - z - z^2 - z^3$ does not have any "simple" roots. I'd really like to factor it into a series of binomials of the form $(1 - cz)$ so that I can apply some of the ordinary generating functions machinery we studied in chapter three.

Being analytically stuck, I turned to Wolfram Alpha which helpfully points out that the full expression has two simple roots:

$$-\sqrt{-1}$$ and $$(-1)^{2/3}$$

I don't know their multiplicity. I thought about writing the expression as:

$$(1 - z(-1)^{-1/3})(1 - z(-1)^{-2/3})$$

But I don't know how to apply generating function machinery to this expresion.

Wolfram Alpha also points out that the Taylor series expansion at 0 is:

$$1 + 2z + 4z^2 + 7z^3 + 13z^4 + O(z^5)$$

Which lead me to try directly computing the Taylor series of the generating function. I didn't see any obvious pattern in the derivatives, but perhaps I missed something. It seemed like I was headed for a huge mess of fractions.

# Conclusion

I'd really like a confirmation that the generating analytic combinatoric expression I derived is correct. I'd also really appreciate some pointers about calculating the series corresponding to my generating function.

Thanks!

• Your work seems fine so far. The roots of the cubic in the denominator are not very nice, but can be found in principle, and then the usual partial fraction methods can be used to get a closed form involving powers of these roots. Your combinatorial analysis implies a very simple recurrence, reminiscent of the Fibonacci recurrence, which can be used to compute the coefficients in the generating function as high as you like. Maybe this isn't the kind of solution you had in mind, however. – Will Orrick Jan 31 '16 at 15:50
• The rational expansion theorem may be what you want, which allows you to do the partial fraction decomposition with $\frac{1 + z + z^2}{1 - z - z^2 - z^3}$. cs.ioc.ee/cm/Rational_Expansion_Theorem.pdf – robit Feb 1 '16 at 7:27

## 1 Answer

First of all your approach is fine and the generating function $G(z)$ is correct. As verification we consider a slightly different approach in the same spirit as it can be found in chapter $5$ in Analysis of Algorithms. Then we look at the coefficients of $G(z)$

Generating function $G(z)$:

We can find in section Bitstrings of chapter $5$ a representation of all bitstrings as \begin{align*} B=\varepsilon+(Z_0+Z_1)\times B \end{align*} which means that each bitstring is either empty or a string starting with $0$ or $1$ followed by a bitstring. A combinatorial construction is \begin{align*} B=SEQ(Z_0+Z_1) \end{align*} meaning a bitstring is a sequence of zero or more occurrences of $0$ or $1$ respectively.

We can characterise general bitstrings also by grouping them into blocks starting with $1$. A bitstring consists of zero or more blocks starting with $1$ and followed by zero or more $0$'s. The bitstring may be preceded by zero or more $0$'s. A combinatorial representation is \begin{align*} B=SEQ(Z_0)SEQ(Z_1\,SEQ(Z_0))\tag{1} \end{align*} The generating function describing the number of bitstrings according to (1) is \begin{align*} B(z)=\frac{1}{1-z}\frac{1}{1-z\frac{1}{1-z}}=\frac{1}{1-2z}=\sum_{n\geq 0}(2z)^n \end{align*} showing us, that we have $2^n$ different bitstrings of length $n$. We can now derive from (1) a representation for all strings which do not contain the substring $000$.

We describe the bitstrings not containing a substring $000$ as a general bitstring with the restriction that each group of $0$'s has maximum length $2$. This implies that we have to exchange in (1) \begin{align*} SEQ(Z_0)=\varepsilon+Z_0+Z_0Z_0+\cdots\qquad\text{with}\qquad \varepsilon+Z_0+Z_0Z_0 \end{align*} We obtain \begin{align*} G=(\varepsilon+Z_0+Z_0Z_0)SEQ(Z_1\,(\varepsilon+Z_0+Z_0Z_0))\tag{2} \end{align*} From (2) we obtain the generating function $G(z)$ as \begin{align*} G(z)&=(1+z+z^2)\frac{1}{1-z(1+z+z^2)}\\ &=\frac{1+z+z^2}{1-z-z^2-z^3}\\ &=1+2z+4z^2+7z^4+13z^5+\cdots \end{align*} which is the same as yours.



Coefficients of $G(z)$:

It's convenient to use the coefficient of operator $[z^n]$ to denote the coefficient of $z^n$ in a generating series.

In this case it is not necessary to derive the coefficients $[z^n]$ in $G(z)$ by finding the roots of the equation $1-z-z^2-z^3$. We instead use a geometric series representation of $G(z)$. Since $1+z+z^2=\frac{1-z^3}{1-z}$ we obtain \begin{align*} G(z)&=\frac{1+z+z^2}{1-z-z^2-z^3}\\ &=\frac{1+z+z^2}{1-z(1+z+z^2)}\\ &=\frac{\frac{1-z^3}{1-z}}{1-z\frac{1-z^3}{1-z}}\\ &=\frac{1-z^3}{1-2z+z^4}\\ &=\frac{1-z^3}{1-(2z-z^4)}\\ &=(1-z^3)\sum_{k\geq 0}(2z-z^4)^k\\ &=(1-z^3)\sum_{k\geq 0}z^k(2-z^3)^k\\ &=(1-z^3)\sum_{k\geq 0}z^k\sum_{l=0}^k\binom{k}{l}(-x^3)^l2^{k-l}\tag{3} \end{align*}

We calculate from (3) the coefficient of $z^n$ of $G(z)$ as

\begin{align*} [z^n]G(z)&=[z^n](1-z^3)\sum_{k\geq 0}z^k\sum_{l=0}^k\binom{k}{l}(-z^3)^l2^{k-l}\\ &=\left([z^n]-[z^{n-3}]\right)\sum_{k\geq 0}z^k\sum_{l=0}^k\binom{k}{l}(-z^3)^l2^{k-l}\tag{4}\\ &=\sum_{k\geq 0}\left([z^{n-k}]-[z^{n-3-k}]\right)\sum_{l=0}^k\binom{k}{l}(-z^3)^l2^{k-l}\\ &=\sum_{{k= 0}\atop{n\equiv k(\text{mod} 3)}}^n\binom{k}{\frac{n-k}{3}}(-z^3)^\frac{n-k}{3}2^{k-\frac{n-k}{3}}\tag{5}\\ &\qquad-\sum_{{k= 0}\atop{n\equiv k(\text{mod} 3)}}^{n}\binom{k}{\frac{n-k}{3}-1}(-1)^{\frac{n-k}{3}-1}2^{k-\frac{n-k}{3}+1}\\ &=\sum_{{k= 0}\atop{n\equiv k(\text{mod} 3)}}^n\left(\binom{k}{\frac{n-k}{3}}+2\binom{k}{\frac{n-k}{3}-1}\right) (-1)^\frac{n-k}{3}2^{k-\frac{n-k}{3}}\tag{6} \end{align*}

Comment:

• In (4) we use the linearity of the coefficient of operator and the rule $[z^{n+k}]=[z^n]x^{-k}$

• In (5) we select the coefficient $l=n-k$ which are congruent $0$ modulo $3$ due to the third power of $z$. We also set the upper limit of the sum to $n$ since we only need coefficients up to $z^n$.

With the formula in (6) we can explicitly calculate the coefficients of $G(z)$.

• @lulu: I've added an answer which might be of interest to you. – Markus Scheuer Feb 1 '16 at 15:11
• Thanks @markus-scheuer! I'm still digesting this answer. How did you move from step 2 to step 3 in the first group of equations under "Coefficients of $G(z)$"? I thought maybe you had used the binomial series, but there's no choose function and the sign on the $-2z -z^4$ changed. – Daniel King Feb 3 '16 at 2:46
• @DanielKing: It's a geometric series rather than a binomial series. I've corrected it and added some more steps in (3) which might be helpful. – Markus Scheuer Feb 3 '16 at 10:04