Inhomogeneous Linear Recurrence relation for Merge Sort confusion on

57 Views Asked by At

Question: $T(n) = 2T(n/2) + n$ ==> Note our result will be $\Theta$(nlog(n))

The very first step you should take is putting everything in the form $T(n)=b^np^n$ so $T(n) - 2T(n/2) = n$

Because this is an inhomogenous relation, we want to turn it into a solvable homogenous problem.


Note we can treat n as 2^k becasue of the bn property.

So now we have $T(2^k) - 2T({2^k}/2) = 2^k$


Next, we multiple both sides by 2 to get the k+1 condition, so $2T(2^k)-4T(2^{k-1}) = 2^{k+1}$


Then we can have a duplicate of the above term, except write it in terms of k+1 so $T(2^{k+1})-2(T^k)= 2^{k+1} $

And then subtract the two terms $T(2^{k+1})-2(T^k)= 2^{k+1} - 2T(2^k)-4T(2^{k-1}) = 2^{k+1}$

to get: $T(2^{k+1})-4T(2^k)-4T(2^{k-1}) = 0$


Next we can rewrite $T(2^k)$ as $X^p$

so,

$X^{p+1} -4X^P -4X^{p-1} = 0$

...

So on $(X^{p-1})(X^2-4X-4)= 0$ //Ignore the first part//

so, $X^2-4X-4=0$ and our roots become x=2,2


Because the roots are not distinct, we use the formula, $\sum c_ir^{p}p^{i-1}$


which brings us to $(2)^p + 2^{p}*(p)$

which $p = 2^k$ and $2^k = n$

and so $2^{2^k}+2^{2^k}(2^k)$ which should be $2^{n}+2^{n}*n$ ...

But my professor treats $\sum c_ir^{p}p^{i-1}$ the $p^{i-1}$ as a separate value of just K not $2^k$ in which then K = log(n) ... why is this?

1

There are 1 best solutions below

0
On

Considering for $n \in \mathbb{N}_{>0}$

$$ T(2^{\log_2 n}) = 2T(2^{\log_2(\frac n2)})+n $$

calling now $T'(u) = T(2^u)$ we have

$$ T'(\log_2 n) = 2T'(\log_2 n - 1) + n $$

and calling now $u = \log_2 n$ we have

$$ T'(u) = 2T'(u-1)+2^u $$

which is a linear recurrence with solution given by

$$ T'(u) = T'_h(u) + T'_p(u)\\ T'_h(u) = 2T'_h(u-1)\\ T'_p(u) = 2T'_p(u-1)+2^u $$

so we have

$$ T'_h(u) = C 2^{u-1} $$

now making $T'_p(u) = C(u) 2^{u-1}$ and substituting into the particular recurrence we obtain the new recurrence

$$ C(u)-C(u-1) = 2\Rightarrow C(u) = 2u $$

so

$$ T'(u) = \left(C+2u \right) 2^{u-1} $$

hence

$$ T(n) = C_1n+n\log_2 n $$