Fast Fourier Transform polynomial multiplication? - algorithm

I am using the FFT to evaluate a polynomial at certain points so that it can be represented using value representation. (representation as a number of points equal to its degree)
However to multiply two polynomials of degree d, I need to evaluate both at 2d + 1 points. However using the FFT for evaluation (multiplying by the dth roots of unity) only evaluates the polynomial at d points. Therefore how can the FFT be used for evaluation for polynomial evaluation if it only evaluates a polynomial at d points? (as opposed to 2d + 1)

You get to choose which n-th roots of -1 you evaluate at. If you need 2d-1 points (as I suspect you do) just use the (2d-1)-th roots of -1. In fact, you would normally use the 2^k-th roots of -1, where 2^k is the first power of 2 >= 2d-1, because it is much easier to get fast FFT for powers of 2. The complexity is still O(d log d) because the definition of O allows for constant factors.

Related

Can one compute the nth Fibonacci number in time O(n) or O(1)? Why?

I asked myself if one can compute the nth Fibonacci number in time O(n) or O(1) and why?
Can someone explain please?
Yes. It is called Binet's Formula, or sometimes, incorrectly, De Moivre's Formula (the real De Moivre's formula is another, but De Moivre did discover Binet's formula before Binet), and involves the golden ratio Phi. The mathematical reasoning behind this (see link) is a bit involved, but doable:
While it is an approximate formula, Fibonacci numbers are integers -- so, once you achieve a high enough precision (depends on n), you can just approximate the number from Binet's formula to the closest integer.
Precision however depends on constants, so you basically have two versions, one with float numbers and one with double precision numbers, with the second also running in constant time, but slightly slower. For large n you will need an arbitrary precision number library, and those have processing times that do depend on the numbers involved; as observed by #MattTimmermans, you'll then probably end up with a O(log^2 n) algorithm. This should happen for large enough values of n that you'd be stuck with a large-number library no matter what (but I'd need to test this to be sure).
Otherwise, the Binet formula is mainly made up of two exponentiations and one division (the three sums and divisions by 2 are probably negligible), while the recursive formula mainly employs function calls and the iterative formula uses a loop. While the first formula is O(1), and the other two are O(n), the actual times are more like a, b n + c and d n + e, with values for a, b, c, d and e that depend on the hardware, compiler, implementation etc. . With a modern CPU it is very likely that a is not too larger than b or d, which means that the O(1) formula should be faster for almost every n. But most implementations of the iterative algorithm start with
if (n < 2) {
return n;
}
which is very likely to be faster for n = 0 and n = 1. I feel confident that Binet's formula is faster for any n beyond the single digits.
Instead of thinking about the recursive method, think of building the sequence from the bottom up, starting at 1+1.
You can also use a matrix m like this:
1 1
1 0
and calculate power n of it. then output m^n[0,0].

Dynamic Programming - "maximize" matrix chain multiplication

I'm now practice dynamic programming by myself. For the classic problem "matrix-chain multiplication" is to find the minimize number of scalar multiplication. Which is,
M[i,j] = 0 if i=j
= Min(i<=k<j){M[i,k-1]+M[k,j]+Pi-1*Pj*Pk}
and its time complexity is O(n^3)
But I'm just curious what if I want to find the "maximization"(instead of min) of scalar multiplication, does it exist a optimal structure and is it possible to solve it in polynomial time?
The exact same reasoning as the minimization applies:
If you multiply a1 ... ai, the resulting matrix dimension does not rely on the internal parenthetization.
It follows that that if the optimal - that is, most expensive - partition of a1 ... ai ... an is to multiply the matrices from 1 to i and from i + 1 to n, then it is composed of the optimal solutions to a1 ... ai and ai + 1 ... an
Since the optimal substructure remains, you can use the same algorithm as minimization (of course, changing the criteria for optimality from minimum to maximum).

Binary XGCD for polynomials

There exists a binary GCD algorithm for finding the greatest common divisor of a number. In general, the GCD can be extended to the XGCD, which can help find a multiplicative inverse in a field.
I am working with binary numbers that represent a polynomial. For example, the bitstring 1101 represents x^3 + x^2 + 1. I need to compute the modular inverse of a random polynomial modulo x^p - 1 for some large known prime p. However, I need to do it in constant time (meaning that the runtime should not depend on the number I am inverting). I know how to make the binary GCD constant time and I know how to implement the XGCD for polynomials in order to compute multiplicative inverses. What I don't know is if there exists a binary GCD equivalent (with corresponding XGCD) for (binary) polynomials?
Yes there is. The "binary" GCD works in any ring where the smallest prime exists. For integers it is 2, hence the name binary. For polynomials, it is x. The algorithm follows the same idea: subtract polynomials to eliminate a free term in one of higher degree, factor out the highest possible power of x, and keep going until the result of subtraction becomes zero.

Optimized running time on two long polynomials' division

In two long polynomials of degree n - 1's division, obviously the reminders with coefficient can be calculated in O(n * n). I want to know if there is any faster algorithms to obtain the reminders, e.g. O(nlogn).
P.S(updated)clarify :if polynomials, p, q have different degree, where deg(p) > deg(q), Is it possible to find the remainders of p / q faster than O((deg(p)- deg(q)) * p) while accuracy don't lose?
If the polynomials are of the same degree, you can determine with one operation the number of times one goes into the other (if the coefficients are integers, this will in general be a rational number). Then you can multiply that fraction times one polynomial and subtract it from the other. This takes O(n) field operations, which is optimal since the answer is of length Θ(n) in the average and worst case.
Yes, there is a faster algorithm, however only for very large degrees. First reverse the polynomials, then use fast (Newton-based) Taylor series division to compute the reversed quotient, reverse the quotient and compute the remainder using fast multiplication.
The first operation has the runtime of a multiplication of polynomials with the degree of the quotient, the second of a multiplication of polynomials of the degree of the divisor.

Polynominal multiplication with even #of coefs. in n distinct multiplications

Looking for some help with an upcoming exam, this is a question from the review. Seeing if someone could restate a) so I might be able to better understand what it is asking.
So it wants me to instead of using extra multiplications maybe obtain some of the terms in the answer (PQ) by subtracting and adding already multiplied terms. Such as Strassen does in his algorithm to compute the product of 2x2 matrices in 7 multiplications instead of 8.
a) Suppose P(x) and Q(x) are two polynomials of (even) size n.
Let P1(x) and P2(x) denote the polynomials of size n/2 determined by the first n/2 and last n/2 coefficients of P(x). Similarly define Q1(x) and Q2(x),
i.e., P = P1 + x^(n/2)P2. and Q = Q1 + x^(n/2) Q2.
Show how the product PQ can be computed using only 3 distinct multiplications of polynomials of size n/2.
b) Briefly explain how the result in a) can be used to design a divide-and-conquer algorithm for multiplying two polynomials of size n (explain what the recursive calls are and what the bootstrap condition is).
c) Analyze the worst-case complexity of algorithm you have given in part b). In particular derive a recurrence formula for W(n) and solve. As usual, to simplify the math, you may assume that n is a power of 2.
Here is a link I found which does polynomial multiplication.
http://algorithm.cs.nthu.edu.tw/~course/Extra_Info/Divide%20and%20Conquer_supplement.pdf
Notice here that if we do polynomial multiplication the way we learned in high school, it would take big-omega(n^2) time. The question wants you to see that there is a more efficient algorithm out there by first preprocessing the polynomials, by dividing it into two pieces. This lecture gives a pretty detailed explanation of how to do this.
Especially, look at page 12 of the link. It shows you explicitly how a 4 multiplication process can be done in 3 when multiplying polynomials.

Resources