Akra–Bazzi method

Template:Short description In computer science, the Akra–Bazzi method, or Akra–Bazzi theorem, is used to analyze the asymptotic behavior of the mathematical recurrences that appear in the analysis of divide and conquer algorithms where the sub-problems have substantially different sizes. It is a generalization of the master theorem for divide-and-conquer recurrences, which assumes that the sub-problems have equal size. It is named after mathematicians Mohamad Akra and Louay Bazzi. A practical variation sometimes referred to as the Ali Najib Variation is commonly used when the exact value of the exponent p satisfying the equation $\alpha a^p+\beta b^p=1$ is difficult to compute analytically.

Formulation

The Akra–Bazzi method applies to recurrence formulas of the form:

$T(x)=g(x)+\sum _{i=1}^k{a}_{i}T(b_{i}x+h_i(x))\text{for }x\ge x_0.$

The conditions for usage are:

• sufficient base cases are provided
• $a_i$ and $b_i$ are constants for all $i$
• $a_i>0$ for all $i$
• $0<b_i<1$ for all $i$
• $|g^{\prime }(x)|\in O(x^c)$, where c is a constant and O notates Big O notation
• $|h_i(x)|\in O\left(\frac{x}{(\log x{)}^2}\right)$ for all $i$
• $x_0$ is a constant

The asymptotic behavior of $T(x)$ is found by determining the value of $p$ for which $\sum _{i=1}^k{a}_i{b}_i^p=1$ and plugging that value into the equation:

$T(x)\in \mathrm{\Theta }\left(x^p(1+\underset{1}{\overset{x}{\int }}\frac{g(u)}{u^{p+1}}du)\right)$

(see Θ). Intuitively, $h_i(x)$ represents a small perturbation in the index of $T$. By noting that $\lfloor b_{i}x\rfloor =b_{i}x+(\lfloor b_{i}x\rfloor -b_{i}x)$ and that the absolute value of $\lfloor b_{i}x\rfloor -b_{i}x$ is always between 0 and 1, $h_i(x)$ can be used to ignore the floor function in the index. Similarly, one can also ignore the ceiling function. For example, $T(n)=n+T\left(\frac{1}{2}n\right)$ and $T(n)=n+T\left(\lfloor \frac{1}{2}n\rfloor \right)$ will, as per the Akra–Bazzi theorem, have the same asymptotic behavior.

Ali Najib variation

When directly solving $\alpha a^p+\beta b^p=1$ for the exponent p is inconvenient, the Ali Najib Variation offers a shortcut. Instead of determining p exactly, one compares the value of $\alpha a^d+\beta b^d$ to 1 for the exponent d appearing in the non-recursive term $g(x)=\mathrm{\Theta }(x^d)$:

If $\alpha a^d+\beta b^d<1$, then $T(n)=\mathrm{\Theta }!\left(n^d\right).$

If $\alpha a^d+\beta b^d=1$, then $T(n)=\mathrm{\Theta }!(n^d\log n).$

If $\alpha a^d+\beta b^d>1$, then $T(n)=\mathrm{\Theta }!\left(n^p\right)$, where p is defined by $\alpha a^p+\beta b^p=1.$

This comparison-based approach is particularly handy when p cannot be expressed in closed form or would require numerical methods.

Example

Suppose $T(n)$ is defined as 1 for integers $0\le n\le 3$ and $n^2+\frac{7}{4}T\left(\lfloor \frac{1}{2}n\rfloor \right)+T\left(\lceil \frac{3}{4}n\rceil \right)$ for integers $n>3$. In applying the Akra–Bazzi method, the first step is to find the value of $p$ for which $\frac{7}{4}{\left(\frac{1}{2}\right)}^p+{\left(\frac{3}{4}\right)}^p=1$. In this example, $p=2$. Then, using the formula, the asymptotic behavior can be determined as follows:

$\begin{array}{rl}T(x)& \in \mathrm{\Theta }\left(x^p(1+\underset{1}{\overset{x}{\int }}\frac{g(u)}{u^{p+1}}du)\right)\\ & =\mathrm{\Theta }\left(x^2(1+\underset{1}{\overset{x}{\int }}\frac{u^2}{u^3}du)\right)\\ & =\mathrm{\Theta }(x^2(1+\ln x))\\ & =\mathrm{\Theta }(x^2\log x).\end{array}$

Significance

The Akra–Bazzi method is more useful than most other techniques for determining asymptotic behavior because it covers such a wide variety of cases. Its primary application is the approximation of the running time of many divide-and-conquer algorithms. For example, in the merge sort, the number of comparisons required in the worst case, which is roughly proportional to its runtime, is given recursively as $T(1)=0$ and

$T(n)=T\left(\lfloor \frac{1}{2}n\rfloor \right)+T\left(\lceil \frac{1}{2}n\rceil \right)+n-1$

for integers $n>0$, and can thus be computed using the Akra–Bazzi method to be $\mathrm{\Theta }(n\log n)$.

See also

• Master theorem (analysis of algorithms)
• Asymptotic complexity

External links

• O Método de Akra-Bazzi na Resolução de Equações de Recorrência Template:In lang