• Abstract and Figures
• Public Full-text

Nonlinear Dyn (2017) 90:2457–2479 DOI 10.1007/s11071-017-3813-6 ORIGINAL PAPER Fractal patterns from the dynamics of combined polynomial root finding methods Krzysztof Gdawiec Received: 14 December 2016 / Accepted: 14 September 2017 / Published online: 22 September 2017 © The Author(s) 2017. This article is an open access publication Abstract Fractal patterns generated in the complex 1 Introduction plane by root finding methods are well known in the literature. In the generation methods of these fractals, Fractals, since their introduction, have been used in arts only one root finding method is used. In this paper, to generate very complex and beautiful patterns. While we propose the use of a combination of root finding fractal patterns are very complex, only a small amount methods in the generation of fractal patterns. We use of information is needed to generate them, e.g. in the three approaches to combine the methods: (1) the use of Iterated Function Systems only information about a different combinations, e.g. affine and s-convex com- finite number of contractive mappings is needed [29]. bination, (2) the use of iteration processes from fixed One of the fractal types widely used in arts is complex point theory, (3) multistep polynomiography. All the fractals, i.e. fractals generated in the complex plane. proposed approaches allow us to obtain new and diverse Mandelbrot and Julia sets together with their varia- fractal patterns that can be used, for instance, as textile tions are examples of this type of fractals [30]. They are or ceramics patterns. Moreover, we study the proposed generated using different techniques, e.g. escape time methods using five different measures: average num- algorithm [34] and layering technique [24]. ber of iterations, convergence area index, generation Another example of complex fractal patterns is pat- time, fractal dimension and Wada measure. The com- terns obtained with the help of root finding methods putational experiments show that the dependence of the (RFM) [18]. These patterns are used to obtain paintings, measures on the parameters used in the methods is in carpet or tapestry designs, sculptures [20]orevenin most cases a non-trivial, complex and non-monotonic animation [22]. For their generation, different methods function. are used. The most obvious method of obtaining various patterns of this type is the use of different root finding Keywords Fractal · Root finding · Iteration · methods. The most popular root finding method used Polynomiography is the Newton method [14,35,37], but other methods are also widely used: Halley’s method [17], the secant method [36], Aitken’s method [38] or even whole fam- ilies of root finding methods, such as Basic Family and Euler-Schröder Family [21]. Another popular method B of generating fractal patterns from root finding methods K. Gdawiec ( ) is the use of different convergence tests [10] and differ- Institute of Computer Science, University of Silesia, B¸edzi´nska 39, 41-200 Sosnowiec, Poland ent orbit traps [7,42]. In [6,34], it was shown that the e-mail: [email protected] colouring method of the patterns also plays a signif- 123 2458 K. Gdawiec icant role in obtaining interesting patterns. Recently, tion of the zeros of complex polynomials, via fractal new methods of obtaining fractal patterns from root and non-fractal images created using the mathematical finding methods were presented. In the first method, the convergence properties of iteration functions. authors used different iteration methods known from It is well known that any polynomial p ∈ C[Z] can fixed point theory [13,23,32]; then, in [11], a pertur- be uniquely defined by its coefficients {an, an−1,..., bation mapping was added to the feedback process. a1, a0}: Finally, in [12] we can find the use of different switch- n n−1 ing processes. p(z) = an z + an−1z +···+a1z + a0 (1) All the above-mentioned methods of fractal genera- tion that use RFMs, except the ones that use the switch- or by its roots {r1, r2,...,rn}: ing processes, use only a single root finding method in the generation process. Using different RFMs, we are p(z) = (z − r1)(z − r2) · ...· (z − rn). (2) able to obtain various patterns; thus, combining them together could further enrich the set of fractal patterns When we talk about root finding methods, the poly- and lead to completely new ones, which we had not nomial is given by its coefficients, but when we want been able to obtain previously. In this paper, we present to generate a polynomiograph, the polynomial can be ways of combining several root finding methods to gen- given in any of the two forms. The advantage of using erate new fractal art patterns. the roots representation is that we are able to change The paper is organised as follows. In Sect. 2,we the shape of the polynomiograph in a predictable way briefly introduce methods of generating fractal patterns by changing the location of the roots. by a single root finding method. Next, in Sect. 3,we In polynomiography, the main element of the gen- introduce three approaches on how to combine root eration algorithm is the root finding method. Many dif- finding methods to generate fractal patterns. The first ferent root finding methods exist in the literature. Let two methods are based on notions from the literature, us recall some of these methods, that later will be used and the third one is a completely new method. Sec- in the examples presented in Sect. 5. tion 4 is devoted to remarks on the implementation of the algorithms on the GPU using shaders. Some graph- – The Newton method [21] ical examples of fractal patterns obtained with the help of the proposed methods are presented in Sect. 5.In p(z) N(z) = z − . (3) Sect. 6, numerical experiments regarding the genera- p(z) tion times, average number of iteration, convergence area, fractal dimension and Wada measure of the gener- – The Halley method [21] ated polynomiographs are presented. Finally, in Sect. 7, we give some concluding remarks. 2p(z)p(z) H(z) = z − . (4) 2p(z)2 − p(z)p(z) 2 Patterns from a single root finding method –TheB4 method—the fourth element of the Basic Family introduced by Kalantari [21] Patterns generated by a single polynomial root finding ( ) method have been known since the 1980s and gained B4 z 6p(z)2 p(z)−3p(z)p(z)2 much attention in the computer graphics community =z − . [28,38,42]. Around 2000, there appeared in the liter- p(z)p(z)2 +6p(z)3 −6p(z)p(z)p(z) ature the term for the images generated with the help (5) of root finding methods. The images were named poly- –TheE method—the third element of the Euler- nomiographs, and the methods for their creation were 3 Schröder Family [21] collectively called polynomiography. These two names were introduced by Kalantari. The precise definition ( ) 2 ( ) that he gave is the following [19]: polynomiography ( ) = ( ) + p z p z . E3 z N z (6) is the art and science of visualisation in approxima- p (z) 2p (z) 123 Fractal patterns from the dynamics 2459 – The Householder method [15] using different metrics, adding weights in the metrics, using various functions that are not metrics, and even p(z)2 p (z) tests that are similar to the tests used in the generation Hh(z) = N(z) − . (7) 2p(z)3 of Mandelbrot and Julia sets were proposed. Examples of the tests are the following: – The Euler–Chebyshev method [39] 2 2 ||zn+1| −|zn| | <ε, (12) 2 2 ( − ) ( ) 2 ( )2 ( ) |0.01(z + − z )|+|0.029|z + | − 0.03|z | | <ε, ( )= −m 3 m p z −m p z p z , n 1 n n 1 n EC z z (8) 2 p (z) 2 p (z)3 (13) |0.04(z + − z )| <ε∨|0.05(z + − z )| <ε, where m ∈ R. n 1 n n 1 n – The Ezzati–Saleki method [9] (14) ( ) where (z), (z) denote the real and imaginary parts ES z 1 4 of z, respectively. = N(z)+ p(N(z)) − . p(z) p(z)+ p(N(z)) In [27], we can find the next modification which later was generalised in [13]. The modification is based on (9) the use, instead of the Picard iteration, of the different In the standard polynomiograph generation algo- iterations from fixed point theory. The authors have rithm, used for instance by Kalantari, we take some used ten different iterations, e.g. area of the complex plane A ⊂ C. Then, for each point – the Ishikawa iteration z0 ∈ A we use the feedback iteration using a chosen root finding method, i.e. zn+1 = (1 − α)zn + αR(un), (15) u = (1 − β)z + β R(z ), zn+1 = R(zn), (10) n n n where R is a root finding method, n = 0, 1,...,M. – the Noor iteration Iteration (10) is often called the Picard iteration. After z + = (1 − α)z + αR(u ), each iteration, we check if the root finding method has n 1 n n = ( − β) + β (v ), converged to a root using the following test: un 1 zn R n (16) vn = (1 − γ)zn + γ R(zn), |zn+1 − zn| <ε, (11) where α ∈ (0, 1], β,γ ∈[0, 1] and R is a root finding where ε>0 is the accuracy of the computations. If method. Moreover, in [13], iteration parameters (α, β, (11) is satisfied, then we stop the iteration process and γ ), which are real numbers, were replaced by complex colour z0 using some colour map according to the iter- numbers.

Load more

  23

Copy Link