CC BY
66
ОБРАБОТКА ИЗОБРАЖЕНИЙ
MATRIX ARITHMETIC BASED ON FIBONACCI MATRICES
Alexey Stakhov (Ukraine) e-mail: anna@nest.vinnica.ua
Introduction
Theory of number systems is one of the most ancient mathematical theories. Historically number systems preceded to origin of number theory and were the first mathematical results influenced essentially on development of number theory and on forming of its basic notions. The discovery of positional principle of number representation considers as one of the greatest mathematical discoveries of the ancient mathematics. The history of the positional number systems began in the Babylonian mathematics. As it is emphasized in [1], "the first of the familiar number systems based on the positional principle was the sexagesimal number system of the ancient Babylonians emerged about 2000 BC".
Let us consider the representation of real number A in the positional number system:
A = Z aiRi
(1)
where at is the numeral of the i-th digit (i = 0, ±1, ±2, ±3, ...) of number system (2); R is a base or a radix of number system (2); R' is the "weight" of i-th digit.
Two basic properties are common for all positional number systems:
(1) each digit takes the values from the finite set M of the permissible values;
(2) the "weight" R' of each digit is the function of its position and of some constant called the base of the number system.
The history of arithmetic [1] shows that the properly positional principle did not change starting since the Babylonian sexagecimal number system. A progress of the positional number systems was connected with a change of their bases and the set M of the numerals used for number representation.
Historically there appeared first the positional number systems with the natural bases (2, 3, 10, 12, 20, 60, etc.). The first step for extension of the functional possibilities of the positional number systems was made by the American scientist Claud Shannon who introduced the number systems with the negative bases [2].
In 1957 the American mathematician G. Bergman introduced the number system with an irrational base
(3). A peculiarity of Bergman's number system consisted of the fact that Bergman used the number t =
1 +
V5
2
called the "golden section" or "golden ratio" as
the base of his number system.
The number system with irrational bases based on the generalized "golden ratios" were introduced by the author of the present article in the 80th [4].
Let us note one peculiarity of the positional representation of (1) and its generalizations given in [3] and [4]. The expression of (1) gives only certain class of real numbers, which could be represented in the form of (1). It is clear that many irrational numbers, for instance the number of n, Euler's number of e cannot be represented in the form of (1). It means that the expression
(1) divides all rational numbers into two parts, the constructive numbers, which could be represented (although potentially) in the form of (1) and the non-constructive numbers, which never (even potentially) cannot be represented in the form of (1).
The works of the Russian scientist Khmelnik [5-9] is of a great interest since point of view of development of theory of number system. He introduced the number systems with complex bases [9] but the main his idea consists of the following. He considered the methods of the positional coding of the complicated mathematical objects so as vectors, matrices, functions, geometric figures [5-8].
Although Khmelnik's representations are complicated for technical realization however his idea about development of the number system theory in the direction of the positional coding of the complicated mathematical objects is very interesting and deserves for a special attention.
The main purpose of the present article is to state the results of investigations of positional matrix representation when the square matrix is represented as a sum of powers of special matrices called the Fibonacci matrices.
1. Fibonacci Q-matrix
In the last decades the theory of the Fibonacci numbers [10] was supplemented by the theory of so-called Fibonacci 0-matrix [10-13]. The latter presents itself the 2x2 matrix of the following form:
Q =
1 1 1 0
Table 1
(2)
Note that the determinant of the Q-matrix equals
to -1.
In the paper [13] devoted to the memory of Verner E. Hoggat, the creator of the Fibonacci Association, it is stated the history of the Q-matrix, given an extensive bibliography on the Q-matrix and on related questions and emphasized the Hoggatt's contribution in development of the Q-matrix theory. Although the name of the " Q-matrix" was introduced before Verner E. Hoggat, just from the paper [11] the idea of the Q-matrix "caught on like wildfire among Fibonacci enthusiasts. Numerous papers have appeared in Fibonacci Quarterly authored by Hoggatt and/or his students and other collaborators where the Q-matrix method became a central tool in the analysis of Fibonacci properties" [13, 250].
Let us take the following theorems for the Q-matrix without a proof [10]:
Theorem 1. For the given integer of n (n = 0, ±1, ±2, ±3, ...) the nth power of Q-matrix is given by
f Fn , i Fn ^
(3)
Qn =
n+1
Fn
F,
n-1 y
where Fn-1 , Fn, Fn+1 are the Fibonacci numbers given with recurrent correlation
Fn = Fn-1 + Fn-2 (4)
for the initial conditions
Fo = 0, F1 = 1. (5)
where n = 0, ±1, ±2, ±3, ... Theorem 2.
Det Qn = (-1)n, (6)
where n is an integer.
From Theorems 1 and 2 there follows one of the fundamental properties connecting neighboring Fibonacci numbers:
Fn+1 Fn-1 - Fn = (-1)n (7)
Let us represent the matrix of (3) in the following
form:
Qn =
( Fn + Fn-1 V Fn-1 + Fn-2
Fn-1 + Fn-2 Fn-2 + Fn-3,
Fn
V Fn-1
F,
n-1
\ f
F
n-2 y
F
n-1 V Fn-2
Fn-2 )
(8)
F
n-3 y
The next theorem follows from (8). Theorem 3.
Qn = Qn-1 + Qn-2. (9)
Let us represent the expression of (9) in the form:
Qn-2= Qn - Qn
(9-a)
The explicit forms of the matrices Qn (n = 0, ±1, ±2, ±3, ..) obtained by means of the recurrent formulas of (9), (9-a) are given in Table 1.
n 0 1 2 3 4
Qn (;:) (:;) (:1) G:) (5 3)
Q-n (0 0) C -') (-3 V)
Theorem 4.
QmxQk = QkxQm = Qm
(10)
3.GeneralizedFibonacci matrices forp-Fibonacci numbers
Let us give some non-negative number p = 0, 1, 2, 3, ... and consider the following recurrent correlation
Fp(n)=Fp(n-1)+Fp(n-p-1) (11)
for the initial conditions
Fp(-p+1)=Fp(-p +2)=...=Fp(0)= 0; Fp(1) = 1(12)
where n = 0, ±1, ±2, ±3, ...
The numerical sequences generated with (11), (12) are called generalized Fibonacci numbers or p-Fibonacci numbers [15].
Note that p-Fibonacci numbers include in itself an infinite number of numerical sequences, in particular, the binary numbers for the case of p=0 and the Fibonacci numbers for the case of p=1.
In [16] the general theory of the Fibonacci matrices based on p-Fibonacci numbers was elaborated. A central notion of this theory is the notion of the Qp-matrix:
f
Qp =
110 0 0 0 10 0 0 0 1
0 0)
00
00
0000 0000 10 0 0
10 01 00
(13)
where the index of p takes the values from the set {0, 1, 2, 3, ..}.
Note that the Qp-matrix is the square (p+1)x(p+1) matrix. It contains the pxp identity matrix bordered by the last row of 0's and the first column, which consists of 0's bordered by 1's. For the cases of p = 0, 1, 2, 3 the Qp-matrices have the following forms, respectively:
Q0 = (1) ;
f 1 1 0^
Q1 =11
11
0
= Q;
Q2 =
0 0 1 100
y
+
Q3 =
( 1
0 0
v1
1 0 0) 010 0 0 1 0 0 0J
Let us take without a proof [16] the following important theorems concerning to the Qp -matrices.
Theorem 5.
(
QP =
Fp (n + 1) Fp (n) Fp(n - P + 1) Fp(n - P)
Fp (n - 1) Fp (n - 2)
Fp (n)
Fp (n - 1)
Fp (n - p + 1)
Fp (n - 2p + 1) Fp(n - p - 1)
Fp (n - p)
A
The formula has a number of (14). Theorem 6.
Det Qnp = (-1)pn, (15)
wherep = 0, 1, 2, 3, ... ; n = 0, ±1, ±2, ±3, ... . Theorem 7.
Qnp = Qnp- + Qnp~ p~\ (16)
Theorem 8.
Qm x Qp=Qp x Qm=Qm+k (17)
It is clear that Theorems 5, 6, 7 and 8 are the generalization of the well-known Theorems 1, 2, 3 and 4 for the classical Fibonacci Q-matrix.
4. Fibonacci matrix representation Let us consider the binary positional presentation of the square (p+1)x(p+1) matrix in the following form:
M = S aiQ i ' p
(18)
where ai is the binary numeral {0, 1}; Qlp is the
weight of i-th digit; Qp is the base of the number system (17); i = 0, ±1, ±2, ±3, ... ;p = 1, 2, 3, ... (the case of p = 0 is eliminated from consideration).
The formula of (18) gives an infinite number of the matrix representations of (18) because each number p generates its proper matrix representation of (18).
Note that the expression of (18) gives a certain class of the square (p+1)x(p+1) matrices, which could be represented in the form of (18). The matrices M, which could be represented (although potentially) in the form of (18) could be called the constructive matrices by analogy with the constructive numbers, which could be represented in the form of (1).
Let us consider now the partial case of (18) corresponding to p=1:
M = S a.Q i
where Q is the Fibonacci Q-matrix of (2).
The abridged notation of the sums of (18) and (19) has the following form:
M = amam-1 ...a1a0 , a-1a-2 ...a-m
(20)
Let us consider the representation of the 2x2 matrices in the code form of (20). It is clear that all powers of the Q-matrix are presented as the following code combinations:
Q0= I = 1, 0; Q1= 10, 0; Q2=100, 0; Q-1= 0, 1;
Q"2= 0, 01; Q"3= 0, 001; ... .
Using (19) one may represent in the code form of (20) all the 2x2 matrices being some sums of the Q-matrix powers. For example the matrix
M = Q3 + Q1 + Q"1 + Q-4 =
2 2J+C 0)+(0 ',)+U -3
6 1
1 5
is represented in the code form of (20) as the following: M = 1010, 1001.
A special interest has the code representations of (20), which results from the following procedure. Let us consider the presentation of the identity matrix in the code form of (20):
M = I = Q = | 0 0 I = 1, 00
(21)
In accordance with the property (9) we can represent the identity matrix of (21) in the following form:
I = Q0 = Q_1 + Q-2 (22)
In fact, using Table 1 we get:
01 1 -1
1 -1 -1 2
10 01
Using (22) we can represent the identity matrix I in the code form:
I = 0,11 (23)
Adding 1 (the identity matrix I) to the 0-th digit of (23) we get the following code combination:
M2 =1,11 (24)
Using (9), we get the new representation of the matrix M2:
M2 = 10, 01 (25)
The code combination of (25) corresponds to the
sum
M2 =Q' +Q~2 =(1 0H.1 2 1=f2 =21 •
+
Then adding 1 (the identity matrix I) to the 0th digit of code combination (25) and using (9), we get the new code combination
M3 = 100, 01, which corresponds to the matrix
2 -2 M3 =Q2+Q 2
2 T -1
1 1J I -1 2
3 0 0 3
= 3I.
Continuing this process, i. e. adding 1 (the identity matrix I) to the 0-th digit of the preceding code combination and using (9), we get the code representations of the following 2x2 matrices :
M1 =I 0 0 0 0 1, 0 0 0 0
M2 = 2I 0 0 0 1 0, 0 1 0 0
M3 = 3I = 0 0 1 0 0, 0 1 0 0
M4 = = 4I 0 0 1 0 1, 0 1 0 0
M5 = = 5I 0 1 0 0 0, 1 0 0 1
M6 = = 6I 0 1 0 1 0, 0 0 0 1
M7 = = 7I 1 0 0 0 0, 0 0 0 1
M 8 = = 8I 1 0 0 0 1, 0 0 0 1
Continuing this process we could represent the matrix
n 0 , Mn = | | = nl
0 n
(26)
in the form of (19). Thus the analyzed procedure of the 2x2 matrix coding permits getting the 2x2 matrices of the kind (26) only and hence the matrix of the kind of (26) are constructive one's regarding to the definition of (19).
Note there exists a one-to-one correspondence between the matrices of (26) regarding to the matrix representation of (19) and natural numbers regarding to Bergman's number system [3]. It means that we can use the isomorphism property for getting of code representation of the 2x2 matrices of the kind of (26). With this aim in view first we have to get the x-code of number n (the x-code is representation of natural number n in Bergman's number system [3]). According to the isomorphism property the x-codes of numbers n coincide with the code representations (20) of the 2x2 matrices of the kind of (26).
Let us consider now the code representation of the 2x2 matrices of the following kind:
M =
a b b a-b
(27)
where a and b are integers.
For getting the code representation of the matrix of (27) we will apply the following procedure. Let us represent the number b as some sum of the Fibonacci numbers, i. e.
b = Fn + Fp + ... + Ft,
where n>p> ...>t..
Next let us form from the number b two new numbers b+ and b- according to the rule:
b+ = Fn+1 + F+ + ... + F+1;
b- = Fn- 1 + Fp-j + ... + Ft-1; Let us note that b = b+ - b. . It follows from the definition (4) for the Fibonacci numbers.
Let us represent now the matrix of (27) as the sum of two matrices:
f a-b+
M=
+V
0
fb b ^ +
b
b
-J
+1a - b+ I 0
0
(a-b)-b-
0
a - b+
Then the matrix
fb
b
\
V b b-J
+ b b_
f F , F ^ Fn+1 Fn
can be represented as the following sum:
V b
I = 1
F Fn-1 J
f F"i Ft >i
V Fn
f F , F ^ Fp+1 FP
Fr
F
p-1
+... +
J
t+1
Ft Ft
t-1 J
= Qn + Qp + _ + Q'.
The code representation of the matrix of
(28)
a - b+ 0
0
a - b_
= (a - b+ ) xI
can be got using the isomorphism between the matrices of the kind of (26) and natural numbers. For that it is necessary to get the x-code of the integer number of (a -b+). The latter representation togethe with the representation of (28) gives a possibility to represent the matrix of the kind of (27) in the form of (19).
For example let us consider the presentation of the matrix
R =
31 15 15 16
in the form of (18). With this aim in view let us represent the number 15 as the sum of two Fibonacci numbers:
15 = F7 + F3 = 13 + 2. Let us form now the numbers
15+ = F8 + F4 = 21 + 3 = 24
and
15- = F6 + F2 = 8 + 1 = 24 and represent the matrix R in the following form:
+
R _
31 15 15 16
21 13 I f 3 2 | f 7 0 13 8 2 1 J + ^ 0 7
Table 2
7 3 7 3 4 -4
= Q7 + Q3 + 71 = Q 7 + Q3 + (Q4 + Q 4; = .
7 5 -4
= Q7 + Q5 + Q 4
Thus the matrix R has the following code presen-
tation:
,31 15 , R = | | = 10100000,0001.
15 16
S _
Let us consider now the matrix
'17 15N 15 2
This one could be represented as the sum:
s = f21 13] + f3 2~l + f-7 0
^13 8) \2 1) \ 0 -7
= Q7 + Q3 -71 = Q7 + Q3 -(q4 + Q"4)
Using (9) we can represent the sum of (29) in the
(29)
0 + 0 _ 0
1 + 0 _ 1
0 + 1_ 1
1 + 1_ 1 1 1 (a)
1 + 1_ 10 0 1 (b)
Using Table 2 for the addition of the matrices S+S 2S, we get the following code representation 2S = 100001001,00101 = 8 3 -3 -5
= Q8 + Q3 + Q 3 + Q 5 =.
34 30 |_ 117 15
30 4 J _ 21 15 2
The identity of (10) underlies the 2x2 matrix multiplication. Table 3 gives the rule of the matrix multiplication.
Table 3
form:
7 3 4 -4 6 4 0 -2
Q7 + Q3 - (Q + Q 4; = Q 6 + Q 4 + Q 0 + Q 2 +
-4 -5 4 -4
+ Q 4 + Q 5 - (Q + Q 4; =
6 3 2 0 -2 -5
= Q6 + Q3 + Q 2 + Q0 + Q 2 + Q 5 =
6 4 0 -2 -5
= Q6 + Q4 + Q0 + Q 2 + Q 5
Thus the matrix S has the following code representation:
(17 15^
S = 1 1 = 1010001,01001.
I15 2)
Above we considered in detail the code representations of the 2x2 matrices in the form of (19). The formula of (18) expands a class of Fibonacci matrix representations and is a source of new investigations in this field.
5. Fibonacci matrix arithmetic
The identities of (9), (10) and their generalizations of (16), (17) form the basis of the Fibonacci matrix arithmetic.
Let us consider the simplest Fibonacci matrix arithmetic corresponding to the case of p=1. Using (9) we can represent the sum of two matrix Q powers in the following form:
Qn + Qn = Qn + Qn-1 + Qn-2 = Qn+1 + Qn-2 . (30)
The identities of (30) underlie the matrix addition. Table 2 gives the addition rule of two single-digit 2x2 matrices.
0 X 0 = = 0
0 X 1 = 0
1 X 0 = 0
1 X 1 1
Let us apply this rule for multiplication of the two matrices
M2 = 10,01 and M2 = 101,01 For multiplication let us represent these matrices in the form with the floating point:
M2 = 1001xQ"2; M4 = 10101xQ-2 After multiplication of the code combinations 1001 x 10101 we get the product
M2 x M4 = 2I x 4I =100010001xQ-4 = = 10001, 0001= Q4 + Q0 + Q-4 =
5 3 3 2
1 0
+ I I + 01
2 - 3 -3 5
= 8I.
Conclusion
The positional matrix representations considered in the present article expand essentially the notion of the mathematical object positional representation. Unlike to the classical positional number systems where the numbers are the objects of code representation the objects of new positional presentation considered in the article are square matrices, which are more complicated mathematical objects in comparison to numbers. The bases of new positional representations are special matrices called the Fibonacci matrices.
It is shown in the article that only the matrices of the special kinds, so called constructive matrices, could be represented using new methods of matrix representation. This fact restricts practical applications of positional representations considered above. However, the
article puts forward a number of theoretical problems concerning to matrix theory, in particular the problem to find a general method of Fibonacci matrix representation for arbitrary square matrix.
The American mathematician George Bergman who is the pioneer in non-traditional number systems wrote in conclusion of his article [3]: "I do not know of any useful application for systems such as this, except as a mental exercise and pastime, though it may be of some service in algebraic number theory".
However progress of computer science refuted this pessimistic Bergman's statement. Just Bergman's number system and its generalization, the "Codes of the Golden Ratios" [4], became of source of modern investigations in the field of the super-fast algorithms of digital signal processing [16].
Probably the matrix representations based on the Fibonacci matrices could bring to new and unexpected applications to digital signal processing.
References
1. Bashmakova, J.G., Youshkevich, A.P. "Origin of Number Systems". Encyclopedia of Elementary Mathematics. Book 1 "Arithmetic". Moscow-Leningrad: Publisher "Gostekhizdat", 1951, 427 p. (In Russian).
2. Shannon C. "A symmetrical notation for numbers". The American Mathematical Monthly, Vol. 57, No 2, 1950.
3. Bergman, G. "A number system with an irrational base". Mathematics Magazine, No 31, 1957, P. 98119.
4. Stakhov, A.P. The Golden Ratio Codes. Moscow: Publisher "Radio and Communication", 1984, 140 p. [In Russian].
5. Khmelnik, S. I. "Coding of the functions". Cybernetics, No 4, 1966. P. 88-92 [In Russian].
6. Khmelnik, S.I. "Coding of the vectors". Cybernetics, No 5, 1969. P. 54-57 [In Russian] .
7. Khmelnik, S.I. "Coding of the plane figures". Automation and Computer Engineering, No 6, 1970. P. 17-21 [In Russian].
8. Khmelnik, S.I. "Algebra of the multi-dimensional vectors, and coding of the space figures". Automation and Computer Engineering, No 1, 1971, pp. 15-19 [In Russian].
9. Pospelov, D.A. Arithmetical Foundations of Digital Computers. Moscow, Publisher "Vyshaja shkola", 1970, 320 p. [In Russian].
10. Hoggat, Verner E. Fibonacci and Lucas Numbers, California: Houghton-Mifflin, Palo Alto, 1969, 92 p.
11. Basin, S.L. Hoggatt, V.E., Jr. "A primer on the Fibonacci sequence, Part II", The Fibonacci Quarterly, No 2, 1963, pp. 61-68.
12. Ivie, John. "A general Q-matrix", The Fibonacci Quarterly, No.3, 1972, pp. 255-261, 264.
13. Gould, H.W. "A history of the Fibonacci Q-matrix and a higher-dimensional problem", The Fibonacci Quarterly, No 1, 1981, pp. 250-257.
14. Stakhov, A.P. "Application of the natural redundancy of the Fibonacci number systems for computer system control", Automation and Computer Engineering, No 6, 1975, pp. 80-87 (In Russian).
15. Stakhov, A.P. Computer Arithmetic based on Fibonacci Numbers and Golden Section: New Information and Arithmetic Computer Foundations, Toronto, SKILLSET Training, 1997, 379 p.
16. Chernov V.M., Pershina M.V. "Fivonacci-Mersenne and Fibonacci-Fermat discrete transforms". Collection of papers "The Golden Section: Theory and Applications" - Boletin de Informatica, No 9/10, 1999.
