Fractals and Chaos _____ Simplified for the Life Sciences Larry S. LiebovitchFractals and Chaos Simplified for the Life Sciences LARRY S. LIEBOVITCH Center for Complex Systems Florida Atlantic Universit ttp:/www.ccs fau.edu/~liebovitch/larry. html New York Oxford OXFORD UNIVERSITY PRESS 1998Contents Part { FRACTALS Introduction 1a 142 113 ‘The Difference between Non-Fractal and Fractal Objects ‘The Sizes of the Features of Non-Fractal and Fractal Objects The Properties of Fractals, Self-Similarity 124 122 123 124 125 126 127 Sealing 137 138 ‘Two Types of Selt-Similarity Examples of Selt Similarity in Space Examples of Self Similarity in Time ‘The Currents through lon Channels Are Self-Similar in Time ‘The Open and Ciosed Times of lon Channels Are Statistically SeltSimilar More Examples of Self-Similarty Biological Implications of Self-Similarty Self-Similarity implies a Scaling Relationship Scaling Relationships Example of a Power Law Scaling ofa Spatial Object: ‘The Length of the Coastine of Britain Examples of Power Law Scalings of Spatial Objects: ‘The Surfaces cf Cell Membranes Example of a Power Law Scaling of a Process in Time: fon Channol Kinetics ‘The Physical Significance of the Scaling Relatonship of lon Channel Kinetics, ‘More Examples of Scaling Relationships Biological Implications of Scaling Relationships xi 12 14 16 18 888 8 & 8 8BContents Dimension 1.41 Dimension: A Quantitative Measure of Solf-Similariy and Scaling 46 1.4.2 The Simplest Fractal Dimension: The SelSimilarity Dimension 48 1.4.3 More General Fractal Dimensions: The Capacity Dimension 50 144 More General Fractal Dimensions: The Hausdortt-Besicovitch Dimer 1.45 Example of Determining the Fractal Dimension: Using the Sotf-Simitarty Dimension 54 1.4.8 Example of Determining the Fractal Dimension: Using the Capacity Dimension and Box Counting 56 1.4.7 Example of Determining the Fractal Dimension: Using the Scaling Relationship 58 1.48 The Topological Dimension 60 1.49. The Embedding Dimension 62 1.4.10 Definition of a Fractal w 1.4.11. Example of the Fractal Dimension: Blood Vessels inthe Lungs 66. 1.4.12. More Examples of the Fractal Dimension 68 1.4.13. Biological Implications of the Frail Dimension 70 Statistical Properties 1.5.1 The Statistical Propertice of Fractals 7 1.52. SelfSimilarty implies That the Moments Do Not Exist 7% 1.5.3. Example Where the Average Doas Not Exist: The St. Potorsburg Game 7% 1.54 Example Where the Average Doos Not Exist: Diffusion Limited Aggregation (DLA) 1.5.5 Example Where the Variance Does Not Exist: ‘The Roughness of Rocks 1.5.6 Example Whore the Variance Doos Not Exist: Electrical Activity in Nerve Cells 1.5.7 Statistical Analysis of the Electrical Activity in Nerve Cells 1.5.8 Example Where the Vatiance Does Not Exist: ‘Blood Fiow in the Heart 1.5.8 Example Where the Variance Does Not Exist: Volumes of Breaths 1.5.10 Example Where the Vatiance Does Not Exist: Evolution 1.5.11 The Distribution of Mutations Is the Same as in the St. Petersburg Game 1.5.12 Statiscal Properties and the Power Spectra 1.5.13 How fo Measure the Propartias of Fractal Data 1.5.14 More Examples ofthe Statistical Properties of Fractals 100 1.5.15 Biological Implications of the Statistical Properties of Fractals 102 1.5.16 Biological Implications of the Statistical Properties of Fractals (continued) 104 s8e8 88 8 88 8 B Fractals and Chaos Simplified for the Life Sciences ‘Summary 161 162 ‘Summary of Fractals ‘Where to Leam More about Fractals Part Il CHAOS Introduction 2.1.1 Two Sets of Data That Look Alike 2.4.2. The Difference between Randomness and Chaos 2.1.3 A Simple Equation Can Produce Complicated Output 21.4 How to Tell Randomness from Chaos 2.1.5 Definiton of Chaos 2.1.6 The Properties of Chaos Phase Space 2.2.1 Phase Space 222 Attractors 2.2.3 The Fractal Dimension of the Attractor Tells Us the Number of Independent Variables Needed to Generate the Time Series 2.2.4 Random is High Dimensional; Chaos is Low Dimensional 2.2.5 Constructing the Phase Space Set from a Sequence of Data Time 2.2.6 Example of Phase Space Sets Constructed from Direct ‘Measurement: Motion of the Surtace of Hair Cells in the Ear 2.2.7 Example of Phase Space Sets Constructed from the Measurementof One Variable 2.2.8 The Beating of Heart Cels 2.2.9 Example of Using the Phase Space Set to Differentiate Random and Deterministic Mechanisms: The Beating of Heart Cells 2.2.10 Overview of the Phase Space Analysis, 2.2.11 The Fractal Dimension ls Not Equel to the Fractal Dimension 2.2.12 Biological Implications of the Phase Space Analysis Sensitivity to Initial Conditions 2.3.1 Lorenz System: Physical Description 2.3.2 Lorenz System: Phase Space Sot 23.3 Lorenz System: Sensitivty to Initial Conditions 2.3.4 Chaos: Deterministic but Not Prodictable 23.5 The Clockwork Universe and The Chaotic Universe 23.6 Lorenz System: A Strange Attractor 23.7. “Strange” and “Chaotic” 23.8 The Shadowing Theorem 2.3.9 Biological Implications of Sensitiviy to Initial Conditions 108 110 116 118 120 122 124 126 130 132 134 136 146 148 150 152 156 158 160 162 164 166 168 1470 172Contents Bifarcations 2.4.1 Bifurcation: An Abrupt Shift in Behavior 2.4.2 Bifurcation Diagrams 24.3 Example of Bifurcations: Theory of Glycolysis 2.4.4 Example of Bifurcations: Experiments of Glycolysis 2.4.5 Example of Bifurcation Diagrams: Theory and Experiments of Giyoolysis 2.4.6 Example of Bifurcations: Motor and Sensory Phase Transitions 2.4.7 Biological Implications of Bifurcation Analyzing Data 25.1 Analyzing Data: The Good News 2.5.2 Example of the Fractal Dimension of the Phase Space Set: Epidemics 2.5.3 Example of the Fractal Dimension of the Phase Space Set: The Heart 2.5.4 Example of the Fractal Dimension of the Phase Space Set: “The Brain 25.5 lon Channel Kinetics: Random or Deterministic? 2.5.6 lon Channel Kinetics: An Interpretation of the Deterministic Model 2.5.7 fon Channel Kinetics: Physical Meaning of the Random and Deterministic Models 2.5.8 Analyzing Data: The Bad News 2.5.9 Some of the Problems of the Phase Space Analysis 2.5.10 Problems: How Much Data? 2.5.11 Problems: The Lag At 2.5.12 Problems: The Embedding Theorems 2.5.13 Problems: The Meaning of a Low Dimensional Attractor 2.5.14 New Methods: Average Direction of the Trajectories 2.5.15 New Methods: Surrogate Data Set 2.5.16 Biological Implications of the Strengths and Weakness in Analyzing Data Control of Chaos 2.6.1. The Advantages of Chaos 2.6.2 Example of Chaotic Control: The Light Output froma Laser 2.6.3 Example of Chaotic Control: The Motion of a Magnetoslastic Ribbon 2.6.4. Biological Implications of the Control of Chaos 2.6.5 Biological Implications of the Control of Chaos (continued) Summary 2.7.1 Summary of Chaos 2.7.2. Where to Lean More about Chaos xiv 176 178 180 182 184 186 188 192 194 196 210 212 214 216 218 240 Fractals and Chaos Simplified for the Life Sciences Part Ill OTHER METHODS The Big Picture 3.1.1 The Big Picture 3.1.2 Some Other Nonlinear Tools xvReferences D. J. Amit. 1989. Modeling Brain Function. Cambridge University Press, New York. D. Avnir, ed. 1969. The Fractal Approach to Heterogeneous Chemistry. John Wiley & Sons, New York. A. Babloyantzand A. Destexhe. 1988. Is the normal heart a periodic oscillator? Biol. Cybern. 58:203-211. A. Babloyantz and A. Destexhe. 1988. The Creutzfeld-Jakob disease in the hierarchy of chaotic attractors. In From Chemical to Biological Organization, M. Markus, S. Muller, and G. Nicolis, eds. Springer-Verlag, New York, pp. 307-316. P. Bakand M. Creutz. 1994. Fractals and self-organized criticality. In Fractals in Science, A. Bunde and S. Havlin, eds. Springer-Verlag, New York, pp. 26- 47. P, Bak, C. Tang, and K. Wiesenfeld. 1988. Scale invariant spatial and temporal fluctuations in complex systems. In Random Fluctuations and Pattern Growth, H. E. Stanley and N. Ostrowsky, eds. Kluwer Academic Pub., Boston, pp. 329-335. M. Bamsley. 1988. Fractals Everywhere. Academic Press, New York. J. B. Bassingthwaighte, L. S. Liebovitch, and B. J. West. 1994. Fractal Physiology. Oxford University Press, New York. J. B. Bassingthwaighte and J. H. G. M. van Beek. 1988. Lightning and the heart: Fractal behavior in cardiac function. Proc. IEEE 76:693-699. K. Y. Billah and R. H. Scanlan. 1991. Resonance, Tacoma Narrows bridge failure, and undergraduate physics textbooks. Am. J. Phys. 59:118-124. A. Block, W. von Bloh, and H. J. Schelinhuber. 1990, Efficient box-counting determination of generalized fractal dimensions. Phys. Rev. A 42:1869- 1874. L. M. Boxt, J. Katz, F.C. Czegledy, L. S. Liebovitch, R. Jones, P. D. Esser, and L, M. Reid. 1994. Fractal analysis of pulmonary arteries: The fractal dimension is lower in pulmonary hypertension. J. Thoracic Imaging 9:8-13. S. R. Brown and C. H. Scholz. 1985. Broad bandwidth study of the topography of natural rock surfaces. J. Geophys. Res. 90:12575-12582. J. Caims, J. Overbaugh, and S. Miller. 1988. The origin of mutants. Nature 335:142-145. E. B. Cargill, H. H. Barrett, R. D. Fiete, M. Ker, D. D. Patton, and G. W. 251References Seeley. 1988. Fractal physiology and nuclear medicine scans. SPIE Medical Imaging 11914:355-361. F, Caserta, H. E. Stanley, W. D. Eldred, G. Daccord, R. E. Hausman, and J. Nittmann. 1990. Physical mechanisms underlying neurite outgrowth: A quantitative analysis of neuronal shape. Phys. Rev. Lett. 64:95-98. A. M. Churilla, W. A. Gottschalke, L. S. Liebovitch, L. Y. Selector, A. T. Todorav, and S. Yeandle. 1996. Membrane potential fluctuations of human. ‘Tlymphocytes have fractal characteristics of fractional Brownian motion. Ann. Biomed. Engr. 24:99-108. M. Ding, C. Grebogi, E. Ott, T. Sauer, and J. A. Yorke. 1993. Plateau onsct for correlation dimension: When does it occur? Phys. Rev. Lett. 70:3872- 3875. W. L. Ditto, S. N. Rausco, and M. L. Spano. 1990. Experimental control of chaos. Phys. Rev, Lett. 65:3211-3214. G. A. Edgar. 1990, Measure, Topology. and Fractal Geometry. Springer-Verlag, New York. S. J. Evans, S. S. Khan, A. Garfinkel, R. M. Kass, A. Albano, and G. A. Diamond. 1989. Is ventricular fibrillation random or chaotic? Circulation Suppl. 80:11-134. F, Family, B. R. Masters, and D. E. Platt. 1989. Fractal pattern formation in human retinal vessels. Physica D38:98-103. J. Feder. 1988, Fractals. Plenum, New York. W. Feller. 1968. An Introduction to Probability Theory and Its Applications, Vol. 1, 3rd. Ed. John Wiley & Sons, New York. N. A. Gershenfeld, 1990. On measuring large dynamical dimensions. Preprint. 1. Giaever and C. R. Keese. 1989. Fractal motion of mammalian cells. Physica D38:128-133. L. Glass, M. R. Guevara, J. Bélair, and A. Shrier. 1984. Global bifurcations of a periodically forced biological oscillator. Phys. Rev. A 29:1348-1357. J. Gleick. 1987. Chaos: Making a New Science. Viking, New York. A. L. Goldberger, V. Bhargava, B. J. West, and A. J. Mandell. 1985. On a mechanism of cardiac electrical stability. Biophys. J. 48:525-528. J. Guckenheimer and P. Holmes. 1983. Nonlinear Oscillations, Dynamical ‘Systems, and Bifurcations of Vector Fields. Springer-Verlag, New York. H. Haken. 1983, Synergetics: An Introduction, 3rd. Ed. Springer-Verlag, New Yori. B. Hess and M. Markus. 1987. Order and chaos in biochemistry. Trends. Biochem. Sci. 12:45-48, A. V. Holden, ed. 1985. Chaos. Princeton University Press, Prince.on, NJ- B. Hoop, H. Kazemi, and L. S. Liebovitch. 1993. Rescaled range analysis of resting respiration. Chaos 3:27-29. XJ. Hou, R. Gilmore, G. B. Mindlin, and H. G. Solari. 1990. An efficient algorithm for fast O(N*In(N)) box counting. Phys. Lett. A151:43-46. P. M. lannaccone and M. Khokha, eds. 1996. Fractal Geometry in Biological 252 Fractals and Chaos Simplified for the Lite Sciences ‘Systems. CRC Press, Boca Raton, FL. K, Kaneko. 1993. Chaotic traveling waves in a coupled map lattice, D68:299-317. une Bip ieee ete Pe Fofaplan and R. J. Cohen, 1990. Is Abrillation chaos? Cire. Res. 67:656- 892. D. T Kaplan and L. Glass. 1992. Direct test for determinism in a time series, Phys, Rev. Lett. 68:427-430. 9. A. S, Kelso, 1995. Dynamical Pattems. MIT Press, Cambridge, MA. D. E. Lea and C. A. Coulson, 1949. The distribution of the number of mutants in bacterial populations. J. Genetics 49:264-285, B. R. Levin, D. M. Gordon, and F. M. Stewart. 1989. Is natural selection the composer as well as the editor of genetic variation? Preprint. L. S. Lieboviteh. 1990. Fractal activity in cell membrane ion channels, In Mathematical Approaches to Cardiac Arrhythmias, ed. J. Jalife. Ann. NY. Acad. Sci. 591:375-391, LS. Lieboviteh, J. Fischbarg, and J. P. Konlarek. 1987. 1on channel kineti A model based on fractal scaling rather than multistate Markov processes. Math. Biosci. 84:37-68, L. S. Lichovitch and T. I. Téth. 1988. A fast algorithm to determine fractal dimensions by box counting. Phys. Lett. A141:386-390. 1. S. Lebovitch and 7. 1. Téth. 1991. A model of ion channel kinetics using deterministic chaotic rather than stochastic processes. J. Theor. Biol 148:243-267. E.N, Lorenz. 1963. Deterministic nonperiodic flow. J. Atmos. Sci. 20:130-141, S- ©. Luria and M. Delbmick. 1945. Mutations of bacteria from virus sensitivity of virus resistance. Genetics 28:491-511. M.A. Mainster. 1990. The fractal properties of retinal vessels: Embryologtcal and clinical implications. Eye 4:235-241, 8. B. Mandelbrot. 1974. A population birth-and-mutation process I: Explicit distributions for the number of mutants in an old culture of bacteria, J. Appl. Prob. 11:437-444. B. B. Mandelbrot. 1983. The Fractal Geometry of Nature. W. H. Freeman and ‘Company, New York. M. Markus, D. Kuschmitz, and B. Hess. 1985. Properties of strange attractors in yeast glycolysis. Biophys. Chem. 22:95-105. ‘ G. Mayer-Kress and S. P. Layne. 1987. Dimensionality of the human clectroencephalogram. In Perspectives in Biological Dynamics and Theoretical Medicine, S. H. Koslow, A. J. Mandel, and M. F. Shlesinger, eds. Ann. NY. Acad. Sci. 504:62-87. P. Meakin. 1986. Computer simulation of growth and aggregation processes. In On Growth and Form: Fractal and Non-Fractal Patterns in Physics, H. E. Stanley and N. Ostrowsky, eds. Martinus Nijhoff, Boston, pp. 111-135. F.C. Moon. 1992. Chaotic and Fractal Dynamics. John Wiley & Sons, New York. 253References E. Mosekilde and L. Mosekilde, eds. 1991. Complexity, Chaos, and Biological Evolution. Plenum, New York. M, A. H. Nerenberg and C. Essex. 1990. Correlation dimension and systematic geometric effects. Phys. Rev. A 42:7065-7074. L. F Olsen and W. M. Schaffer. 1990. Chaos versus noisy periodicity: Alternative hypotheses for childhood epidemics. Science 249:499-504. A. R. Osborne and A. Provenzale. 1989. Finite correlation dimension for stochastic systems with power-law spectra. Physica D35:357-381. E. Ott. 1993. Chaos in Dynamical Systems. Cambridge University Press, New York. D. Paumgariner, G. Losa, and E. R. Weibel. 1981. Resolution effect on the stereological estimation of surface and volume and its interpretation in terms of fractal dimensions. J. Micros. 121:51-63. C.-K. Peng, S. V. Buldyrev, A. L. Goldberger, S, Havlin, F. Sciortino, M. Simons, and H. E. Stanley. 1992. Long-range correlations in nucleotide sequences. Nature 356:168-170. P. E, Rapp, T. R. Bashore, J. M. Martinerie, A. M. Albano, I. D. Zimmerman, and A. I. Mees. 1989. Dynamics of brain electrical activity. Brain Tomography 2:99-118. L. F. Richardson. 1961. The problem of contiguity: An appendix to statistics of deadly quarrels. General Systems Yearbook 6:139-187. R. Roy. T. W. Murphy Jr. T. D. Maier, and Z. Gills. 1992. Dynamical control of a chaotic laser: Experimental stabilization of a globally coupled system. Phys. Rev. Lett. 68:1259-1262. R. H. Scanlan and J. W. Vellozzi. 1980. Catastrophic and annoying responses of long-span bridges to wind action. In Long Span Bridges, E. Cohen and B. Birdsall, eds. Ann. N.Y. Acad. Sci. 352:247-263. W. M. Schaffer and M. Kot. 1986. Differential systems in ecology and epidemiology. In Chaos, A. V. Holden, ed. Princeton University Press, Princeton, NJ. C. A. Skarda and W. J. Freeman. 1987. How brains make chaos in order to make sense out of the world. Behav. Brain Sci. 10:161-195. L. A. Smith. 1988. Intrinsic limits of dimension calculations. Phys. Lett. A133:283-288, T. G. Smith Jr, W. B, Marks, G. D. Lange, W. H. Sheriff, Jr., and E. A. Neale. 1988. A fractal analysis of cell images. J. Neurosci. Meth. 27:173-180. F, Takens. 1981. Detecting strange attractors in turbulence. In Dynamical ‘Systems and Turbulence, D. A. Rand and L.-S. Young, eds. Springer-Verlag, New York, pp. 366-381. M. C. Teich. 1989, Fractal character of the auditory neural spike train. IEEE Trans. Biomed. Engr. 46:41-52. M. C. Teich, D. H. Johnson, A. R. Kumar, and R. G. Turcott. 1990. Rate fluctuations and fractional power-law noise recorded from cells in the lower auditory pathway of the cat. Hear. Res. 46:41-52. 254 Fractals and Chaos Simplified for the Life Sciences M. C. Teich, S. M. Khana, and S. E. Keilson. 1989. Nonlinear dynamics of cellular vibrations in the organ of Corti. Acta Otolaryngol (Stockholm) Suppl. 467:265-279. J. Theiler, S. Eubank, A. Longtin, B. Garldrikian, and J. D. Farmer. 1992. Testing for nonlinearity in time series: The method of surrogate data. Physica D58:77-94. B. J. West and A. L. Goldberger. 1987. Physiology in fractal dimensions. Am. Sct, 75:354-36' S. Wolfram. 1994. Cellular Automata and Complexity. Addison-Wesley, Reading, MA. A. Wolf, J. B. Swift, H. L. Swinney, and J. A. Vastano. 1985. Determining Lyapunov exponents from a time series. Physica D16:285-317. N. Xu and J. Xu. 1988. The fractal dimension of the EEG as a physical measure of conscious brain activities. Bull. Math. Biol. 50:559-565. J.P. Zbilut, G. Mayer-Kress, P, A. Sobotka, M. O'Toole, and J. X. Thomas, Jr, 1989. Bifurcations and intrinsic chaos and 1/f dynamics in an isolated perfused rat heart. Biol. Cybern. 61:371-381. 255Index action potentials 84-87, 144-147 air 156 airways 14, 22, 40, 62 Albano 197, 199 Amit 249 anti-persistence 90 art 173, 189 arteries 14, 22, 66-70, 88 atmosphere 156 ATP 180-185 attractors definition 132, 133 dimension 134-137, 142, 143 logistic 123, 127. 134, 135, 142, 143 Lorenz 131-135, 158, 159, 168, 167 strange 132, 158, 166-169 auditory nerve cells 84-87, 140, 141 average (see mean, sample) average directions in phase space 218, 219 Avnir 111 Babloyantz 197, 199 bacterial colonies 68 bad news 206 Bak 249 balls 50, 52, 56 Barnsley 111 Barrett 97 Bashore 199 Bassingthwaighte 89, 99, 111, 241 Belair 145, 147 Bernoulli Daniel 78 Niklaus 78 Bhargava 97 bifurcation diagram 178, 179, 184, 185 bifurcations (see chaos, bifurcations) Billah 203 biological implications chaos analyzing data 222 bifureations 188 control of chaos 232, 234 phase space 152 sensitivity to initial conditions 172 fractals dimension 70 scaling 42 self-similarity 24 statistical properties 102, 104 Birdsall 203 Block 57 blood flow 88, 89, 100 blood vessels body 22, 68 heart 88, 89 lungs 66-68 retina 14, 15, 68, 70 Boca Raton, FL 126 bone 68 books (sce fractals, references, suggested) 259(see chaos, references, suggested) Boston, MA 130 box counting dimension 56, 57 Boxt 67 brain 198 breaths 22, 90, 91, 100 Britain 32, 33, 58 - Brown 83 Bunde 249 butterfly effect 126 Cairns 93 calculus 206 cancer 104 capacity dimension 47, 50-52, 56 capillaries 68 Cargill 97 Caserta 15 Cauchy 206 cell cycle 104 cells auditory 84-87, 140, 141 cancer 104 cultured 14-17, 22, 144-147 cycle time 104 heart 144-147 membrane 22, 34, 35, 40, 68 motion 16, 17, 22, 140, 141 nerve 14, 15, 22, 68, 84-87, 100 T-lymphocytes 16, 17. 100 vestibular 86, 87 voltage 16, 17, 22, 84-87, 144- 147 cellular automata 248, 249 central limit theorem 86 chaos analyzing data (Part 2, Chapter 5) 191-223 bad news 206, 222 biological implications 222 good news 192, 222 definition 118, 124, 168 Index 260 bibliography 240, 241 bifurcations (Part 2. Chapter 4) 175-189 bifurcation diagram 178 biological implications 188 definition 176 deterministic vs. random mechanisms 188 glycolysis 180-185 logistic equation 176-179 motor and sensory phase transitions 186, 187 period doubling 184 biological examples 140-141, 144-147, 180-187, 194-205 biological implications analyzing data 222 bifurcations 188 control of chaos 232, 234 motor phase transitions 186, 187 phase space 152 sensitivity to initial conditions 172 sensory phase transitions 186 control of chaos (Part 2, Chapter 6) 225-235 iological implications 232-5 definition 226 magnetoelastic ribbon 230 laser 228, 229 deterministic vs. random bifurcations 188, 192 ton channels 200-205 new methods 218-221 number of independent variables 124, 134-137 phase space 134-137, 192 vectors 218, 219 problems with the analysis 206-217, 222 surrogate data sets 220, 221 Fractals and Chaos Simplified for the Life Sciences dimension (see chaos, phase space, dimension) introduction (Part 2, Chapter 1) 115-127 phase space (Part 2, Chapter 2) 129-158 biological implications 152 construction 122, 123, 130, 131, 188, 189, 142, 143 definition 130 deterministic vs. random 122, 126, 134-137, 142, 148 dimension 122, 126, 134- 137, 142, 150, 192 electrocardiogram 196, 197 electroencephalogram 198, 199 epidemics 194, 195 hair cells 140, 141 heart cells 144-147 logistic equation 122, 126, 134 Lorenz equation 131, 132, 134, 158, 166 overview of the phase space analysis 148, 149 problems with the phase space analysis 206-217, 222 random mechanism 122, 126, 136 vectors 218, 219 problems with the chaos analysis amount of data 210 embedding theorems 214 interpretation of low dimensional attractor 216 lag 212 summary 208 properties 124, 126, 238 references, suggested 240, 241 2, Chapter 3) 155-173 biological implications 172 definition 162 Lorenz equations 156-161, 166, 167 Poincaré 164 time series 160, 161 strange 166-169 summary 238 chaotic universe 164 characteristic scale 6 checkerboard 80, 81 chemical reactions 40, 68, 180- 185 chickenpox 194, 195 Churilla 17 clockwork universe 164 coastline 32, 33 Cohen 197, 203 coin games ordinary coin toss 78 St. Petersburg Game 78 complex output 120, 124 construction of the phase space set 122, 130, 138, 142 control of chaos (see chaos. control) convection 156 convulsions 198 comnea 68 correlations 70, 220 Coulson 95 coupled maps 248, 249 covering dimension 47, 60, 61 Creutzfeld-Jakob 199 Creutz 249 cultured cells 14-17, 22, 144.147 Czegledy 67 Daceord 15 data (see time series) definitions: bifurcation 176 chaos 118, 124, 168 sensitivity to initial conditions (Part —_—_control of chaos 226 261dimension 46 fractal 64 phase space 130 sealing 28 self-similarity 12 sensitivity to initial conditions 162 statistical properties of fractals 74 strange 168 Delbruck 93 Destexhe 197, 199 deterministic but not predictable 162-165 mechanisms 116-126 phase space 192, 126, 136, Ms vs. random 116-119, 122, 126, 186, 149, 148, 152. 188, 192 ion channels 200-205 problems with the analysis 206-217, 222 surrogate data sets 220-221 vectors in phase space 218, 219 Diamond 197 diffusion limited aggregation (DLA) 70, 80, 81 dimension (Part I, Chapter 4) 45- a biological implications 70 blood vessels in the lungs 66, 67 box counting 56, 57 capacity 47, 50-52, 56 covering 47, 60, 61 definition 46 deterministic mechanism 122, 126, 136, 142, 148 embedding 46, 62, 138, 142 examples 68 fractal biological fractals 66-69 Index biological chaos 140, 141, 146, 194-199 definition 64, 65 phase space 122, 126, 184,-137, 142, 148, 150, 168 types 46-59 Hausdorff-Besicovitch 47,52,53, iterative 47, 60. 61 logistic equation 122. 123, 127. 134-137 Lorenz equations 134, 135, 158 not equal to the dimension 150, 151 number of independent variables 134 phase space 122, 126, 136, 138, 142. 148 problems with the phase space analysis 206-217, 222 random mechanism 122. 126. 136, 142, 148, 150, 168, 194-199, 210-217, 222 scaling 58 self-similarity 47-49, 54 topological 46, 47. 60, 61, 64 Ding 211 distributions fractal 70, 84, 102. 104 Gaussian 70, 84, 86, 102, 104 Poisson 86 probability density function (pdf 12, 20, 21 pulse number 84-87 stable 70, 102, 104 Ditto 231 DNA 100 dynamical systems 124, 240 ear 84-87, 140, 141 Edgar 111 Eldred 15 262 Fractals and Chaos Simplified for the Life Sciences clectrocardiogram 196, 197 electroencephalogram 198, 199 electromagnetic radiation 188 embedding 122, 130, 138, 142 embedding dimension 46, 62, 138, 142 embedding theorems 208, 214 energy levels in proteins 22 enzymes 68, 180-185 falso, see proteins) epidemics 194, 195 epilepsy 198 Esser 67 Essex 211 Evans 197 evolution 92, 94, 100 eye 14, 15, 62, 70 failure of the Tacoma Narrows Bridge 202, 203 Falke 19 Family 15 Fano factor 86, 87, 98 Feder 111 Feller 79 fibrillation 196 finger 186, 187 Fourier 220 fractal dimension (see dimension, fractal) fractal dimension is not equal to the fractal dimension 150, 151 fractals definition 64 dimension (Part 1, Chapter 4) 45-71 biological implications 70 ‘blood vessels in the lungs 66, 67 box counting 56. 57 capacity 47, 50, 52, 56 covering 47, 60, 61 definition 46 embedding 46, 47, 62 263 examples 68 fractal 46-58, 64-70 Hausdorf-Besicovitch 47, 52, 53 iterative 47, 60, 61 sealing 58 self-similarity 47-49, 54 topological 46, 60, 64 bibliography 110, 111 iological examples fractal dimension 68 scaling 41 self-similarity 22 statistical properties 100 biological implications dimension 70 sealing 42 self-similarity 24 statistical properties 102, 104 introduction (Part 1, Chapter 1) 3-10 Properties 8, 108 references, suggested 110, 111 scaling (Part 1, Chapter 3) 27 43 iological implications 42 coastline 32, 33 definition 28 examples 32, 34, 36, 40 from self-similarity 28 fon channel Kinetics 36-39 mean 98 membrane area 34, 35 power laws 28, 30 self-similarity (Part 1, Chapter 2) 11-95 alrways in the lung 14, 15 biological implications 24, 70 blood vessels in the retina 14, 15 cell motion 16. 17 cell voltage 16, 17definition 12 examples 22 geometric 12 implies that the mean does not exist 76 fon channels 18, 19, 20, 21 nerve cells 14, 15 statistical 12 statistical properties (Part 1, Chapter 5) 73-105 biological implications 102, 104 blood flow in the heart 88, 89 breath volume 90, 91 checkerboard 80, 81 definition 74 diffusion limited aggregation (LA) 80, 81 evolution 92, 93, 94, 95 examples 100 mean does not exist 74-81, 84, 86, 96, 98, 102, 104 nerve cells 84, 85, 86, 87 rocks 82, 83 St. Petersburg Game 78, 79, 94, 95 variance does not exist 76, 82-98, 102, 104 summary 108 Freeman 199 future work 102, 214 gambling 78 Garfinkel 197 Gaussian 70, 84, 86, 102, 104 geometric self-similarity 12 Gershenfeld 211 Giaever 17 Gilles 19 Glass 145, 147, 219 Gleick 241 Gliwice, Poland 126 index glycolysis 168-185 Goldberger 15, 97 good news 192 Gordon 93 Gottschalke 17 Guckenheimer 241 Guevara 145, 147 H90 hair cell 84-87, 140, 141 Haken 186, 187 Hausdorff-Besicovitch dimension 47, 52, 53 Hausman 15 Havin 249 hearing 84-87, 140, 141 heart 22, 68, 88, 97, 100, 144- 147, 196 herpes simplex 68 Hess 181, 183 heterodyne optical interferometry 140 Holden 241 Holmes 241 homeostatis 234 Hoop 91, 99 Hou 57 Hurst exponent 90, 91 Hurst rescaled range analysis 90, 91, 98, 99 Tannaccone 111 immune system 232 implications, biological (see biological implications) independent variables (see variables) initial conditions 160-163, 166- 173 interferometry 140 intestine 22 intracellular voltage 16, 17, 22, 84-87, 144-147 introductions chaos (Part 2, Chapter 1) 237- 264 Fractals and Chaos Simplified for the Life Sciences 241 fractals (Part 1, Chapter 2) 3-9 other methods 246, 247 ion channels 18-22, 36-39, 62, 68 deterministic model 200-205 iterative map model 200-205 Markov model 200, 201, 204, 205 random vs. deterministic 200- 205 iterative dimension 47, 60, 61 iterative map 200, 201 Johnson 85, 87, 99 Jones 67 Kaneko 249 Kaplan 197, 219 Katz 67 Kazemi 91 Keese 17 Kelso 187 Ker 97 Khan 197 Khanna 87 Khokha 111 kinetics ion channels 18-22, 36-39, 62. 68 washout 40 Koch curve 54, 55, 60, 64 Kot 195 Kumar 85, 87, 99 lag chaos analysis 138, 139, 150, 151 problems 208, 212-215 Hurst analysis 90, 91, 98, 99 Lange 15 Laplace 164 laser 228, 229 Layne 199 Lea 95 Leibniz. 206 Levin 93 Liapunov exponent 160 Liebovitch 17, 21, 37, 57, 91. 111. 201, 241 limits 50, 52, 74, 78-90, 102 linear 246, 247 Upids 68 liver 22, 68, 97, 100 local thermodynamic equilibrium 204 logistic equation 116-123, 176-179 Lorenz equations 131, 133, 134, 135, 156-161, 166, 234 Losa 35 Jungs airways 14, 15, 22, 40, 62 blood vessels 66, 67 Larla 93 magnetoelastic ribbon 230, 231 Mainster 15 Mandelbrot 95, 111 Mandell 97 Markov model 200, 201, 204, 205 Marks 15 Markus 181, 183, 185, 199 Martinerie 109 Masters 15 mathematical tools 246, 247 Mayer-Kress 197, 199 Meakin 81, 99 mean does not exist 74-81, 84, 86, 96, 98, 102, 104 population 74, 75, 78 sample 74, 78, 80, 64 scaling 98 self-similarity 76 similar in random and deterministic 116, 117 St. Petersburg Game 78, 79 mean squared deviation 82, 98 mechanisms 116 265Index measles 194, 195 Overbaugh 93 measure theory 110 oxygen 66 mechanisms deterministic (see deterministic, parameter 176, 222 mechanisms) Patton 97 random (see random, mechanisms) Paumgartner 35 meditation 198 Peng 23 Mees 199 period doubling bifurcation 184 membrane 22, 34, 35, 40, 68 persistence 90 metronome 186 Miller 93 Misler 19 moments (see mean and variance) ‘Moon 241 Mosekilde 241 motor phase transition 186, 187 Muller 199 mutations 92, 94, 100 Neale 15 Nerenber 211 neural networks 248, 249 neurons 14, 15, 22, 68, 84-87, 100 Newton 206, 246 Nicolis 199 Nittman 15 nonlinear 246, 247 nonlinear dynamics 240 nonstationary 104 number of independent variables (see variables, number) O'Toole 197 one-over-f noise 96, 97 optical interferometry 140 Ostrowsky 81 other methods 246-249 cellular automata 248, 249 coupled maps 248, 249 neural networks 248, 249 self-organizing critical systems 248, 249 Ott 241 phase space (see chaos, phase space) phase space set (see attractors) phase transitions 186, 187 placenta 22 Platt 15 Poincaré 164 Poisson distribution 86 population mean 74, 75, 78 potential energy function 186 power spectrum, chaos 116 power law 96, 97 predictability 162-165 probability density function (see distributions) problems with the chaos analysis amount of data 210 embedding theorems 214 interpretation of low dimensional attractor 216 lag 212 summary 208 Projects to do 102, 214 proteins energy levels 22 ion channel 18-22, 36-39, 62, 68 surface 40, 68 vibrations 68 pulse number distribution 84-87 R/S 90, 91, 98, 99 radioisotope tracer 68, 97, 100 random mechanisms 116, 118, 122, 266 Fractals and Chaos Simplified for the Life Sciences 126 phase space 122, 126, 136 vs. deterministic 116-119, 122, 126, 136, 142, 148, 152, 188, 192 ion channels 200-205 problems with the analysis 206-217, 222 surrogate data sets 220-221 vectors in phase space 218, 219 Rapp 199 Rauseo 231 reaction rates 40, 68, 180-185 references suggested chaos 240, 241 fractals 110, 111 other methods 248, 249 Reid 67 relative dispersion 88, 89, 98 resonance 202 retina 14, 15, 62, 70 Richardson 33 rocks 82 root mean square (RMS) 82 Roy 229 sample mean 74. 78, 80, 84 scaling, fractals (see fractals, scaling) Scanian 203 Schaffer 195 Scholz 63 Seeley 97 Selector 17 self-orgenization 186 self-organizing critical systems 248, 249 self-similarity dimension (see dimension, self-similarity) self-similarity, fractals (see fractals, self-similarity) sensitivity to initial conditions (see chaos. sensitivity to initial conditions) sensory phase transition 186 shadowing theorem 170, 171 ‘Sheriff 15 Shrier 145, 147 significant digits 162 Skarda 199 Smith 15, 211 Sobotka 197 ‘Spano 231 St. Petersburg Game 78, 79. 94. 95 stable 234 standard deviation 98 Stanley 15, 81, 99 stationary 104 statistical properties (see fractals, statistical properties) statistical self-similarity 12 statistical tests 102 Stewart 93 strange attractor 182, 158, 166-169 substances in the blood 40 sugar 180-185, summaries chaos (Part 1, Chapter 7) 237- 24) fractals (Part 1, Chapter 6) 107-111 other methods 248, 249 surrogate data set 220, 221 synergetics 186 T-lymphocytes 16, 17, 100 Tacoma Narrows Bridge 202, 203 Takens’ theorem 138, 214, 215 teeth 68 ‘Teich 85, 87, 99, 141 Theiler 221 theorem central limit 86 embedding 122, 130, 138, 142 shadowing 170, 171 ‘Takens' 138, 214, 215 thermodynamic equilibrium 204 267‘Thomas 197 time series construction of phase space set 122, 180, 138 deterministic 116-119, 122, 126, 136 dimension 150, 151 logistic 116-123, 126, 127 Lorenz 131, 135, 160, 161 random 116-119, 122, 126, 136 ‘Todorov 17 topological dimension 46, 47, 60, 61, 64 topology 110 Toth 57, 201 trajectory 170, 218 transient 132 Turcott 85, 87, 99 ‘universe chaotic 164 clockwork 164 unstable 234 variability, biological 172, 226, 232 variables number 124 related to the dimension 134- 139, 142, 148 variance blood flow in the heart 88, 89 breath volume 90, 91 does not exist 76, 82-98, 102, 104 evolution 92, 98, 94, 95 Index Hurst rescaled range analysis 90, 91, 98 mean squared deviation 98 mutations (see variance, evolution) nerve cells 84, 85, 86, 87 relative dispersion 88, 89, 98 rocks 82 root mean square deviation 82 similar in random and deterministic mechanisms 116 vectors in phase space 218, 219 veins 14, 22, 66-70, 88 ‘Vellozi 203 vestibular nerve cell 86, 87 vibrations in proteins 68 volume of breaths 22, 90, 91, 100 washout kinetics 40 Weibel 35 West 15, 97, 111, 241 Westerman 87 winnings 78 Wolff 211 Wolfram 249 world 246, 247 x-rays 66, 67, 68 Xu 199 Yeandle 17 Zbilut 197 Zimmerman 199 268Fractals and chaos are currently generating excitement across various scientific and medical disciplines. Biomedical investigators, graduate students, and undergraduates are becoming increasingly interested in applying fractals and chaos (nonlinear dynamics) to a variety of problems in biology and medicine. This accessible text lucidly explains these concepts and illustrates their uses with examples from biomedical research, The author presents the material in a very uniquc, straightforward manner that avoids technical jar- gon and does not assume a strong background in mathematics, The text uses a step-by-step approach by explaining one concept at a time in a set of facing pages, with text on the left page and graphics on the right page; the graphics pages can be copied directly onto transparencies for teaching. Ideal for courses in biosta- tistics, fractals, mathematical modeling of biological systems, and related courses in medicine, biology, and applied mathematics, Fractals and Chaos Simplified for the Life Sciences will also serve as a useful resource for scientists in biomedicine, physics, chemistry, and engineering. ABOUT THE AUTHOR Larry S. Liebovitch is Professor at the Center for Complex Systems and Brain Sciences, the Center for Molecular Biology and Biotechnology, and the Department of Psychology at Florida Atlantic University. He is the coauthor, with James Bassingthwaighte and Bruce West, of Fractal Physiology (OUP, 1994). | i 9 apissti2024 Cover DESIGN Eb ATKESON/BERG DESIGN ISBN 0-19-512024-8Part | FRACTALS Objects or processes whose small pieces resemble the whole. 1 Introduction 3 2 Self-Similarity u 3 Scaling 27 4 Dimension 45 5 Statistical Properties 73 6 Summary 1071.2.6 FRACTALS / Self-Similarity More Examples of Self-Similarity 1. Self-Similarity in Space We have already seen that there are self-similar patterns in the branching of the arms (dendrites) of nerve cells, the arteries and the veins in the retina, and the tubes that bring air into the lungs. Additional examples of seli-similarity in space are the branching structures of the conduction fibers in the heart (His-Purkinje system) that spread the wave of muscle contraction during the heartbeat; the tubes (ducts) in the liver that bring the bile to the gallbladder; and the arteries and the veins throughout the body. Many surfaces in the body have self-similar undulations with ever finer fingers or pockets. These ever finer structures increase the area available for the exchange of nutrients, gasses, and ions. These surfaces include the lining of the intestine, the boundary of the placenta, and the membranes of cells. These properties are self-similar in physical space. There are also self- similar properties in more abstract spaces such as the energy space used to characterize molecules. For example, the distribution of the spacing between energy levels in some proteins is self-similar. 2. Self-Similarity in Time We have already seen that there are self-similar patterns in time in the motion of the heights of cells above the dish in which they are growing, in the variation of the voltage across the cell membrane, and in the timing of the switches between the open and closed states of an ion protein. Additional examples of self-similarity in time include the electrical signal generated by the contraction of the heart and the volumes of breaths over time drawn into the lung. 22 Fractals and Chaos Simplified for the Life Sciences 1.26 Biological Examples of Self-Similarity spatial dendrites of neurons airways in the lung ducts in the liver vessels in the circulatory system intestine placenta cell membrane energy levels in proteins temporal fluctuations in the heights of cell above their substrate voltage across the cell membrane timing of the opening and closing of ion channels heartbeat volumes of breaths 231.2.7 FRACTALS / Self-Similarity Biological Implications of Self-Similarity How can the human genome with only 100,000 genes contain all the information required to construct structures like the heart, which has 1,000,000 capillaries, or the brain, which has 100,000,000,000 nerve cells? Perhaps the genes do not determine these structures. They may instead determine the rules that generate these structures. Repeated application of these rules may then lead to self-similar structures with many pieces at different resolutions. 24 Fractals and Chaos Simplified for the Lite Sciences How is the body formed? 100,000 genes DNA 1,000,000 capillaries 0 2 100,000,000,000 nerve cells => | fures > DNA Self-Similar Structures Repeated Application of these Rules144 FRACTALS / Introduction Fractals and Chaos Simplified tor the Life Sciences 144 The Difference between Non-Fractal and Fractal Objects Fractal Object: Non-Fractal 1. Non-Fractal As a non-fractal object is magnified, no new features are revealed. 2. Fractal As a fractal object is magnified, ever finer features are revealed. ‘The shapes of the smaller features are kind of like the shapes of the larger features.11.2 FRACTALS / Introduction The Sizes of the Features of Non-Fractal and Fractal Objects 1. Non-Fractal The size of the smallest feature of a non-fractal object is called its characteristic scale. ‘When we measure the length, area, and volume at a resolution that is finer than the characteristic scale, then all of the features of the object are included. Thus the measurements at this resolution determine the correct values of the Jength, area, and volume. 2, Fractal A fractal object has features over a broad range of sizes. ‘As we measure the length, area, and volume at ever finer resolution, we include ever more of its finer features. Thus the length, area, and volume depend on the resolution used to make the measurement. Fractals and Chaos Simplified for the Life Sciences 112 Non-Fractal Size of Features e? ton ese 1 characteristic scale Ce € @ Fractal Size of Features 2em icon 120m _ 140m - many different scales14.3 FRACTALS / introduction The Properties of Fractals 4. Self-Similarity A coastline looks wiggly. You would think that as you enlarge a piece of the coastline the wiggles would be resolved and the coastline would look smooth. But it doesn't. No matter how much you enlarge the coastline it still looks just as wiggly. The coastline is similar to itself at different magnifications. This 1s called self-similarity. 2. Scaling Because of self-similarity, features at one spatial resolution are related to features at other spatial resolutions. The smaller features are smaller copies of the larger features. The length measured at finer resolution will be longer because it includes these finer features. How the measured properties depend on the resolution used to make the measurement is called the scaling relationship. 3. Dimension ‘The dimension gives a quantitative measure of self-similarity and scaling. It tells us how many new pieces of an object are revealed as it is viewed at higher magnification. 4. Statistical Properties ‘Most likely, the statistics that you were taught in school was limited to the statistics of non-fractal objects. Fractals have different statistical properties that may surprise you! Fractals and Chaos Simplified for the Lite Sciences 14.3 Self-Similarity Scaling The value measured for a property depends on the resolution at which it is measured. Dimension Statistical Properties moments may be zero or non-finite. forexample, mean — O variance > =1.24 FRACTALS / Self-Similarity Two Types of Self-Similarity 1. Geometrical Self-Similarity In geometrical self-similarity little pieces of an object are exact smaller copies of the whole object. ‘The little pieces are geometrically similar to the whole object. This is usually true only for mathematically defined objects. 2, Statistical Self-Similarity ‘The little pieces of real biological specimens are usually not exact smaller copies of the whole object. Thus, they cannot be geometrically self-similar. However, the little pieces of real biological specimens can be kind-of-like their larger pieces. The statistical properties of the little pieces can be geometrically similar to the statistical properties of the big pieces. This is called statistical self-similarity. For example, the statistical property could be the length of the perimeter of an organ. Statistical self-similarity means that the length measured at one Fesolution is geometrically similar, that is, proportional to the Jength measured at other resolutions. If Q(x) is the length measured at resolution r, and Qlar) 4s the length measured at resolution ar, then Q(ar) = k Q(r) where kis the constant of proportionality. The statistical properties of an object are described by the number of pieces of each size that make up the object. The function that tells how many pieces of each size that make up the object is called the probability density function (pdf). The formal mathematical definition of statistical self-similarity is that the probability density function (pdf measured at resolution ris geometrically similar, that is, has the same shape, as the probability density function (pdf) measured at resolution ar. We use the shorter expression “self-similarity” to denote the statistical self-similarity of biological objects in space or processes in time. This means that the smaller pieces are like the larger pieces but they are not exact copies of the larger pieces. 12 Fractals and Chaos Simplified for the Life Sciences 1.24 Self-Similarity Geometrical The magnitied piece of an object is an exact copy of the whole object. Statistical The value of statistical property Q(r) measured at resolution r, is proportional to the value Q(ar) measured at resolution ar. Q(ar) = k Q(r) paf [afar] £ paf (ka(9] a 131.2.2 FRACTALS / Selt-Similarity Examples of Self-Similarity in Space Examples of statistical self-similarity in space are: 1. The branching patterns of the arms (dendrites) of nerve cells growing in the retina in the eye or growing in a nutrient medium in a plastic culture dish, 2. The arteries and veins in the retina, and 3. The tubes that bring air into the lungs. ‘These objects are self-similar because the pattern of the branching of the large structures is repeated in the branching of the ever smaller structures. ‘The fractal analysis of these branching pattems sheds light on how the dendrites of the nerve cell grow and how the airways in the hung are formed in the embryo. Determining the fractal properties of the blood vessels in the retina may be useful in diagnosing diseases of the eye or in determining the severity of the disease. ‘The analysis of nerve cells and blood vessels in the retina and in the nutrient medium of a culture dish is simplified by the fact that the retina and the nutrient media are both very thin. Thus the branching pattern is 2- dimensional. Analysis of a photograph of a 3-dimensional pattern would have to compensate for the images of the branches projected on top of each other. 14 Fractals and Chaos Simplified for the Life Sciences 122 Branching Patterns nerve cells in the retina, and in culture Caserta, Stanley, Eldred, Daccord, Hausman, and Nittmann 1990 Phys. Rev. Lett. 64:95-98 Smith Jr, Marks, Lange, Sheriff Jr., and Neale 1989 J. Neurosci. Meth. 27:173-180 Z — blood vessels airways in the retina in the lungs Family, Masters, and Platt 1989 West and Goldberger 1987 Physica D38:98-103 Am. Sci. 75:354-365 Mainster 1990 Eye 4:235-241 Lx 15123 FRACTALS / Self-Similarity Examples of Self-Similarity in Time Examples of statistical self-similarity in time are: 1, The average height of a layer of cells above the plastic culture dish in which they are growing, and 2, The electrical voltage across the cell membrane of a T-lymphocyte cell. ‘These measurements are self-similar because the pattern of the smaller fluctuations over brief times is repeated in the larger fluctuations over longer times. 16 Fractals and Chaos Simplified for the Life Sciences 1.23 Variations in Time cell height above a substrate “ Giaeverand Keese 1989 Physica D36:128-133 ara acoutsmon {ANB PROCESSING. mV (a measure of the average height) i voltage across the cell membrane ™" Churilla, Gottschalk, Liebovitch, Selector, Todorov, and Yeandle 1996 Ann. Biomed. Engr. 24:99-108 s mV124 FRACTALS / Self-Similarity Fractals and Chaos Simplified for the Life Sciences 124 The Currents through lon Channels Are Self-Similar in Time Currents Through lon Channels Jon channels are proteins in the cell membrane with a central hole that sap : ‘; allows ions such as sodium. potassium, and chloride to get into or out of the ATP sensitive potassium channel in cell. The structure of the ion channel protein can change, closing this hole and B cell from the pancreas blocking the flow of ions. The small electrical current (10°12 amp) due to these ions can be measured in an individual ion charinel molecule. This current is high when the channel is open and low when it Is closed. Gilles, Falke, and Misler (Liebovitch 1990 Ann. N.Y. Acad. Sci. 591:375-391) When a recording of current is played back at low time resolution, the times during which the channel was open and closed can be seen. When one of these open or closed times is played back at higher time resolution, it can be seen to consist of many briefer open and closed times. ‘The current through the channel is self'similarin time because the pattern f, = 10 Hz of open and closed times found at low time resolution is repeated in the open and closed times found at higher time resolution. 50 msec 19125 FRACTALS / Self-Similarity The Open and Closed Times of lon Channels Are Statistically Self-Similar ‘The number of times that the ion channel is closed for each Tange of closed times is called the closed time histogram. We recorded the current through potassium channels in the cells that line the inside of the comea in the eye. We played back these current recordings through an analog to digital (A/D) converter into a computer. We then used the computer to determine the histogram of the closed times at these different time resolutions. Sampling the data at high time resolution by using a high A/D rate best reveals the statistical properties of the brief closed times. Sampling the data at a low time resolution by using a low A/D rate best reveals the statistical properties of the long closed times. ‘The horizontal time axis is different for histograms computed from the data sampled at different A/D rates. However, these histograms all have the same shape. These statistical distributions of closed times measured at different resolutions in time are geometrically similar. ‘Thus the closed times are statistically self-similar. 20 Fractals and Chaos Simplified for the Life Sciences 125 Closed Time Histograms potassium channel in the comeal endothelium Liebovitch et al. 1987 Math. Biosci. 84:37-68 100 4a Alp = 5000 Hz 1 mM 20 100 180 260 340 100 10 AlD= 1700 Hz 1 a 66 330 504 858 1122 100 we AD =500 Hz 1 1 200 1000 "800 2600 3400 400 8 AD = 170 Hz 1 7 60 ‘3300 ‘5840 ase) 1220 Closed Time In ms 24FRACTALS Scaling 13.5 13.6 13.7 1.3.8 Self-Similarity Implies a Scaling Relationship Scaling Relationships Example of a Power Law Scaling of a Spatial Object: The Length of the Coastline of Britain Examples of Power Law Scalings of Spatial Objects: The Surfaces of Cell Membranes Example of a Power Law Scaling of a Process in Time: lon Channel Kinetics ‘The Physical Significance of the Scaling Relationship of lon Channel Kinetics More Examples of Scaling Relationships Biological Implications of Scaling Relationships 271.34 FRACTALS / Scaling Self-Similarity Implies a Scaling Relationship ‘Smaller pieces of a fractal will be seen at finer resolution. A measurement made at finer resolution will include more of these smaller pieces. Thus the value measured of a property, such 2s length, surface, or volume, will depend. on the resolution used to make the measurement. How a measured property depends on the resolution used to make the measurement is called the scaling relationship. Sclf-similarity specifics how the small picces are related to the large pieces. ‘Thus self-similarity determines the scaling relationship. ‘The mathematical form of self-similarity determines the mathematical form of the scaling relationship. The mathematical form of self-similarity is that the value Q(ar) of a property measured at resolution ar is proportional to the value Q(t) measured at resolution r. That ts, Qfar)= k Q(r), where k isa constant, From this form, it can be shown that the scaling relationship has one of two possible forms. 1. Power Law The simplest form of the scaling relationship is that the measured value of a property Q{r) depends on the resolution used to make the measurement with the equation: Q(t) =B r®. In this equation B and bare constants. This form is called a power law. 2. Full Form The full form of the scaling relationship is the equation: Qt) = B r? {Logir]/Logial) where B, b, and a are constants and fix) is a periodic function such that f(1+x)=flx). 28 Fractals and Chaos Simplified for the Life Sciences 1.3.4 Self-Similarity <> Scaling Q(wy=BP Gen =k Qt) =B P HLogte/Lostal) Self-Similarity can be satisfied by the power law scaling: Q(n=Brb Proof: Using the scaling relationship to evaluate Q(r) and Q(ar), Q(r=B rb Q(ar) = Bab ;b f k= ab then Q(ar) =k Q(r) Self-Similarity can be satisfied by the more complex scaling: logr Q() = Brbt( —2_ ) where (14x) = f00 ) (ega) Proof: Using the scaling relationship to evaluate Q(a) and Q(ar), ayer t eee) Q (a) =Bab PEE) = Bab by OOS Oar) =Babih(14 7085) =Babrbi( 285) f k=a> then = Qar)= kK Q(t) 291.3.2 FRACTALS / Scaling Fractals and Chaos Simplified for the Life Sciences 1.3.2 Sealing Relationships Scaling Relationships most common form: ‘The scaling relationship describes how the measured value of a property =Brd Ql) depends on the resolution r used to make the measurement. It can Power Law Q(n=Br Q(r) Log Q(r) 1. Power Law & “2 Se ‘The sunplest and most common form of the scaling relationship 1s that re £o Q(t) = B®, where B and b are constants. 2 £5 = £a On a plot of the logarithm of the measured property, Log [Q(z], versus g a g the logarithm of the resolution used ta make the measurement, Log [x], this a De scaling relationship is a straight Hine. g 3 é £ ‘Such power law scaling relationships are characteristic of fractals. £ a Power law relationships are found so often because so many things in nature are fractal. ogr ‘ Logarithm of the resolution used tome Ke resolution used to make the measurement 2. Full Form ‘The full form of the scaling relationship is that less common, more general form: Ql) = Br? {1 + Log{r)/Log{al] where B, b, and a are constants and ffx) is a periodic function such that f{1+4x)=fx) Q()=Brht eet ) On a plot of the logarithm of the measured property, Log [Q(t)]. versus the Qn Log Q(r) logarithm of the resolution woed to make the measurement, Log [r], this scaling relationship is a straight line with a periodic wiggle. measurement Logarithm of the measurement r Logr resolution used to make the Logarithm of the resolution used to Measurement make the measurement 30 311.3.3 FRACTALS / Scaling Example of a Power Law Scaling of a Spatial Object: The Length of the Coastline of Britain Richardson measured the length of the coastline of Britain by laying small straight line segments of the same length, end to end, along the coastline. The length of these line segments set the spatial resolution of the measurement. The total length of the coastline was the combined length of all these line segments When he measured the coastline at finer resolution, the smaller line segments included the smaller bays and peninsulas that were not included in the measurement at coarser resolution. These additional smaller bays and peninsulas increased the total length of the coastline. He found that the length of the coastline was ever longer as he measured it at ever finer resolutions. He plotted the logarithm of the total length of the coastline, Log [Qtr)]. versus the logarithm of the length of the line segments used to do the measurement, Log [r|. The data were a straight line. He found that Log Q(x) was proportional to (-1/4) Log {r]. This means that the scaling relationship of the measurement of the Igngth of the coastline has the power law form that Q(1) is proportional to r~ 1/4), 32 Logjg (Total Length in Km) Fractals and Chaos Simplified for the Lite Sciences 1.3.3 How Long is the Coastline of Britain? Richardson 1961 The problem of contiguity: An Appendix to Statistics of Deadly Quarrels General Systems Yearbook 6:139-187. AUSTRALY ap IAN Cons; —~] 40 be = s+ cIRCLE SOUTH AFRICAN COAST SERMAN, TAND-FRONTIER, ftH_2 2 Wer, co, AST OF oping in as Logo (Length of Line Segments in Km) 33134 FRACTALS / Scaling Fractals and Chaos Simplified for the Life Sciences 1.3.4 Examples of Power Law Scalings of Spatial Objects: Scal ing of The Surfaces of Cell Membranes Membrane Area ‘Membranes encase living cells, as well as form different compartments wages te om ile! cope Peng om end et 16 such membranes from photographs taken with an electron microscope. The a 4 magnification of the electron microscope set the spatial resolution of the measurement of the surface area. imi inner mitochondrial membrane As the magnification was increased, then more undulations in the membranes were detected, and the amount of membrane surface measured omi outer mitochondrial membrane increased. : . er endoplasmic reticulum They found that the surface area Q(t) measured at resolution r of the endoplasmic reticulum, the outer mitochondrial membrane, and the inner mitochondrial membrane each had a power law scaling of the form that Q(r) 30 was proportional to 2, where the value of b was different for each type of 30 membrane. 20 This power law scaling of the membrane surface area Q(t) with the g resolution r used to measure it appears as a straight line on a plot of 2 “ Log [Qirll_ versus Log [r]. The slope of these lines, which is equal to the =. 7s value of b, is different for each type of membrane. > 3 3° = a ace a6 i ee 6B z | 2 Be | a fe & ow 2 < 072 g g ag s g | \ Sver.cy a 2 & ft4 4 4 ' 23828 g 8 + E888 8 2 1 4 6 8 10 20 40 60 10° RESOLUTION SCALE n=a mM” 34 3513.5 FRACTALS / Scaling Example of a Power Law Scaling of a Process in Time: lon Channel Kinetics We wanted to find out if there is a scaling relationship present in the open and closed times that we recorded from a potassium ion channel in the cells that line the inside of the cornea in the eye. ‘We needed to determine how the probability to change from closed to open depends on the time resolution used to measure it. The channel must be closed long enough for us to see it as closed. Thus, from our data, we determined the conditional probability that a closed channel will open, given that it has already remained closed for a certain amount of time. That certain amount of time then sets the time resolution at which the measurement is done. We called this conditional probability the effective kinetic rate constant, ker and the time resolution at which it was measured the effective time scale, te; We plotted the logarithm of the effective kinetic rate constant Log [kegd versus the logarithm of the effective time scale Log {tggd at which it was measured. ‘The data were a straight line on this plot. ‘That tells us that the ion channel kinetics has a power law scaling characteristic of a fractal, where kegr is proportional to tes? 36 Fractals and Chaos Simplified for the Life Sciences 1.3.5 Scaling of lon Channel Kinetics Liebovitch et al. 1987 Math. Biosci. 84:37-68 70 pS Channel, on cell, Corneal Endothelium 1000 Koq in Hz effective 100 ¥ kinetic rate constant | 104 id Ip a oy 1 10 100 1000 toe in msec effective time scale 3713.6 FRACTALS / Scaling The Physical Significance of the Scaling of lon Channel Kinetics The scaling relaticnship in the timing of the switching between the open and closed states could be interpreted in two different ways: 1. It could reflect the energy structure of the ion channel protein. The channel protein could exist in a large number of similar, but not identical, shapes. In this interpretation, the scaling relationship can be used to determine the heights of the energy barriers that separate the closed shapes from their corresponding open shapes. 2. It could reflect the dymamies of how the shape of the ion channel protein changes in time. In this interpretation, the scaling relationship can be used to determine how the energy barrier between the closed state and the open state varies in time. ‘The fractal approach focused attention on the dynamics, that is, the motions of the structures within ion channel proteins, on how past motions affect future behavior, and on the importance of the large number of slightly different shapes of the channel protein. The results of 20 years of experiments on other proteins also suggested that internal motions and the large numbers of possible protein shapes play an important role in how all proteins work. In contrast, most scientists who studied only ion channels had analyzed their data as if ion channel proteins had only a few states with fixed structures. Our work made some of these scientists very unhappy. Scientists, like anyone else, sometimes react with hostility to new ideas that challenge their beliefs and their previous work. 38 Fractals and Chaos Simplified for the Life Sciences 1.3.6 Two Interpretations of the Fractal Scalings Structural Energy Dynamical ig tp Hi closed time Energy open 39 The scaling relationship reflects the distribution of the activation energy barriers between the open and closed sets of conformational substates. The scaling relationship reflects the time dependence of the activation energy barrier between the open and closed states.137 FRACTALS / Scaling More Examples of Scaling Relationships 1. Scaling Relationships in Space We have already seen that there is a scaling relationship that tells us how the amount of membrane area measured in a cell depends on the resolution in space used to measure it. Additional examples of scaling relationships in space include how the diameter of the tubes that bring air into the lung, the width of the spaces between the cells that line the capillaries in the lung, and the surface areas of proteins each depend on the spatial resolution used to measure them. 2. Scaling Relationships in Time ‘We have already seen that there is a scaling relationship that tells us how the rate of switching between the open and closed states of an ion channel protein depends on the resolution in time used to measure it Another example of scaling relationships in time is the rate of chemical reactions that are limited by the time it takes for the molecules to reach each other. The reaction slows down with time because the nearby molecules have reacted and it takes ever longer for the unreacted molecules to reach each other. The scaling relationship describes how the kinetic rate constant of the reaction depends on the time measured since the reaction began. Asscaling relationship in time is also present in the washout kinetics of how the concentration of substances in the blood decay with the time measured since the substances were first injected into the blood. 40 Fractals and Chaos Simplified for the Life Sciences 1.3.7 Biological Examples of Scaling Relationships spatial area of endoplasmic reticulum membrane area of inner mitochondrial membrane area of outer mitochondrial membrane diameter of airways in the lung size of spaces between endothelial cells in the lung surface area of proteins temporal kinetics of ion channels reaction rates of chemical reactions limited by diffusion washout kinetics of substances in the blood at138 FRACTALS / Scaling Biological Implications of Scaling Relationships 1. No Unique, “Correct” Value for a Measurement ‘The scaling relationship tells us how the value measured for a property, such as length, area, or volume, depends on the resolution used to make the measurement. There is no one value that represents the “correct” value of the property being measured. The value measured for a property depends on the resolution used to make the measurement. 2. Measurements Made at Different Resolutions Will Be Different The value measured for a property depends on the resolution used to make the measurement. Thus measurements made at different resolutions will yield different values. This means that the differences between the values measured by different people could be due to the fact that each person measured the property at a different resolution. It also means that it is very important to state the resolution at which a measurement is performed. 3. Importance of the Scaling Relationship Our method of analyzing the data must be consistent with the characteristics of the data if the results are to be meaningful. Only if the analysis properly characterizes the data can it give us clues about the nature of the process that produced the data. The measurement of the value of a Property at only one resolution is not useful to characterize fractal objects or processes. Instead, we need to determine how the values measured for a Property depend on the resolution used to make the measurement, namely, the scaling relationship. This change, from measuring a single value to measuring how the values depend on the resolution is called a change in paradigm. 42 Fractals and Chaos Simplified for the Life Sciences 13.8 Scaling Logarithm of the measurement Resolution <> Perimeter = 8cm icm Resolution ~_ Perimeter = 12cm 05cm one measurement: scaling relationshi not so interesting much more interesting @ ~~ one value Logarithm of the measurement y~<— slope Logarithm of the resolution used to make the measurement Logarithm of the resolution used to make the measurement 43Babb FRACTALS Dimension Dimension: A Quantitative Measure of Self-Similarity and Scaling The Simplest Fractal Dimension: The Self-Similarity Dimension More General Fractal Dimensions: The Capacity Dimension More General Fractal Dimensions: The Hausdorff-Besicovitch Dimension 1.4.5 Example of Determining the Fractal Dimension: Using the Self-Similarity Dimension 1.4.6 Example of Determining the Fractal Dimension: Using the Capacity Dimension and Box Counting 1.4.7 Example of Determining the Fractal Dimension: Using the Scaling Relationship The Topological Dimension The Embedding ension Definition of a Fractal Example of the Fractal Dimension: Blood Vessels in the Lungs More Examples of the Fractal Dimension Biological Implications of the Fractal Dimension 451.44 FRACTALS / Dimension Dimension: A Quantitative Measure of Self-Similarity and Scaling The dimension gives a quantitative measure of the fractal properties of self- similarity and scaling. The dimension tells us how many additional smaller pieces of an object are revealed when it is magnified by a certain amount. ‘There are a number of different dimensions. Each dimension measures different properties of an object. They can be grouped into three different types of dimensions: 1. Fractal Dimension ‘The fractal dimension describes how an object fills up space. It gives us information about the length, area, or volume of an object. Its value can be an integer or a fraction (for example: 2 or 1.7) 2. Topological Dimension The topological dimension describes how points within an object are connected to each other. It tells us that an object is an edge, a suriace, or a solid. Its value is always an integer (for example: 1 or 2). 3. Embedding Dimension ‘The embedding dimension describes the space that contains the object. It tells us that this space is a line, an area, or a volume, Its value can be an integer or a fraction (for example: 2 or 1.7). 46 Fractals and Chaos Simplified for the Life Sciences 1.4.1 Dimension A quantitative measure of self-similarity and scaling. The dimension tells us how many new pieces we see when we look at finer resolution. Fractal Dimension space filling properties of an object e.g. self-similarity dimension capacity dimension Hausdorff-Besicovitch dimension Topological Dimension how points within an object are connected e.g. covering dimension iterative dimension Embedding Dimension the space that contains an object 4714.2 FRACTALS / Dimension The Simplest Fractal Dimension: The Self-Similarity Dimension The fractal dimension describes the space filling properties of an object. ‘There are many different fractal dimensions. Each one characterizes the space filing properties of an object in a slightly different way. The simplest fractal Gimension is called the self-similarity dimension. Consider a geometrically self-similar fractal object made up of line segments. To evaluate the self-similarity dimension we divide each line segment into Mi smaller line segments. This will produce N smaller objects. If the object is geometrically self-similar, each of these smaller objects is an exact but reduced size copy of the whole object. The self-similarity dimension. @ is then found from the equation N= M®, This equation can also be written as d = Log (N) / Log (M). For example, when M = 3, we replace cach line segment of an object with 3 little line segments. If we divide a line into thirds, it produces 3 little lines. Thus N = 3, The equation N= M4 has the form 3= 34. Hence, the self-similarity dimension dof the line is equal to 1. If we divide each side of a square into thirds, it produces 9 little squares. Thus N= 9. The equation N= M® has theform 9 = 3%. Hence, the self- similarity dimension d of the square is equal to 2. If we divide each side of a cube into thirds, it produces 27 little cubes. ‘Thus N = 27. The equation N= M4 has the form 27 = 3%. Hence, the self- similarity dimension d of the cube is equal to 3. These dimensions computed from the self-similarity dimension are consistent with our intuitive idea that the dimensions of length, area, and volume should be 1, 2, and 3. 48 Fractals and Chaos Simplified for the Life Sciences 1.4.2 Fractal Dimensions Self-Similarity Dimension N_ new pieces when d each line segment =- M is divided by M. N ‘ 3-3! _—_—_— > t 2 2 =d=1 2 | 2 9 = 4914.3 FRACTALS / Dimension More General Fractal Dimensions: The Capacity Dimension ‘The self-similarity dimension requires that each little object formed by dividing the line segments of the whole object into smaller pieces must be an exact copy of the whole object. Thus the self-similarity dimension can only be used to analyze objects that are geometrically self-similar. To determine the dimension of irregularly shaped objects requires a more general form of the fractal dimension. Two such forms are the capacity and the Hausdorff Besicovitch dimension, To evaluate the capacity of an object we cover it with “balls” of a certain radius 5. We find the smallest number of balls N(c) needed to cover all the parts of an object. We then shrink the radius of the balls and again count the smallest number needed to cover the object. The capacity is the value of Log N(x) / Log (1/r) in the limit as the radius of the balls shrinks to 0. ‘The capacity is a generalization of the self-similarity dimension. The self- similarity dimension d = Log N/ Log M, where N is the number of smaller copies of the whole object seen when each line segment is replaced by M pieces. The spatial resolution in the self-similarity dimension is proportional to 1/M. The spatial resolution in the capacity is proportional to the radius r of the balls used to cover the object. Thus r is proportional to 1/M. To arrive at the capacity from the self-similarity dimension, we first replace M by 1/r. ‘This leads us to d = Log N/ Log (1/r), Second, instead of counting N, the number of smaller copies of the whole object, we count N(r) the number of balls needed to cover the object. This leads us to d = Log Nr) / Log(1/r). Last, we take the limit of Log Nir) / Log(1/r) as the radius of the balls shrinks to 0. 50 Fractals and Chaos Simplified for the Life Sciences 143 Fractal Dimensions Capacity Dimension N(r)_ balls of radius r needed to cover the object. Relationship to self-similarity dimension: M=1/r, then N=Mo 511.4.4 FRACTALS / Dimension More General Fractal Dimensions: The Hausdorff-Besicovitch Dimension The Hausdorfl-Besicovitch dimension is usually what mathematicians mean when they say “fractal dimension,” The formal definition of the Hausdorff-Besicovitch dimension is quite technical. Only a brief hint of the those details will be given here. ‘The Hausdorff-Besicovitch dimension is similar, but not identical to, the Sapacity. In the capacity. we count the number of balls N(x) of a given radius Tae ded to cover an object. The equation d = Log Nir) / Log (1/1), implies Tae aPaciy dimension d is determined directly from how the number of balls needed to cover the object varies with the radius of the balls, In the Hausdorff-Besicovitch dimension, we cover the object with sets. We evaluate the sum of a function applied to the diameter of exch covering set. ‘This function is called the gauge function, Analogous to the capacity and sai, ity dimension, the gauge function that 1s used 1s to raise the diameter. ¢ of each set to the power s. The sum of the diameters of all the sets each Taised to the power 5s is then computed. The behavior of this sum as a function of s is then studied in the limit as the diameter r of the sets approaches 0. As r approaches 0, this sum will grow very large if s is less than a certain number, and it will grow very small if sis greater thane certain number. The value of the number thal separates these two types of behavior is called the Hausdorf-Besicovitch dimension. We can now see how the Hausdorff-Besicoviteh dimension is similar to the Will grow very large if sd 5 =x = wn = i 53145 FRACTALS / Dimension Example of Determining the Fractal Dimension: Using the Self-Similarity Dimension qhe Koch curve is constructed by starting with an equilateral triangle. At each stage in the construction, each line segment is divided into thirds. The two end pleces are left intact. Each middle piece is then replaced by two Pieces. Each of the 4 pieces is 1/3 as long as the original line segment.” ‘This Procedure can be repeated forever. Each original side of the triangle is 3 units long. After the first stage of fonsiruction each side is 4 units long. At each stage in the construction, the Jength of the perimeter of the Koch curve increases by 4/3. When there is an infinite number of stages in the construction, then the perimeter of the Koch curve is infinitely Jong. The perimeter of the Koch curve is geometrically self similar. At each stage in the construction the original segment was a straight line, and the new smaller segments are also straight lines. Thus the smaller pieces are geometrically similar to the original segment. We can use the self-similarity dimension to determine the dimension of the perimeter of the Koch curve. As the spatial resolution is increased by a factor of 3, we see 4 new pieces. apatis. when the size of the lines making up the perimeter is reduced by 1/3 of their original length, then we find 4 small pieces. The self-similarity dimension d_ is the logarithm of the number of new pieces divided by the logarithm of the factor of the reduction in size of each pice. Thus d= Log (4) / Log (3 54 Fractals and Chaos Simplified for the Life Sciences 1.4.5 Fractal Dimension of the Perimeter of the Koch Curve aa When we look at 3x finer resolution, we see 4 additional smaller pieces. Self-Similarity Dimension: Log (number of new pieces) — Log 4 42645 7 Log (factor of finer resolution) Log 3 55146 FRACTALS / Dimension Example of Determining the Fractal Dimension: Using the Capacity Dimension and Box Counting ‘The capacity dimension d = Log Nir) / Log (1/1), in the limit where r approaches 0, where Nir) is the smallest number of balls of radius r needed to cover an object. A useful way to evaluate the capacity is to use “balls” that are the boxes of a rectangular coordinate grid. This method is called box counting. For example, we cover an object with a grid and count how many boxes of the grid contain at Jeast some part of the object. We then repeat this measurement a number of times, each time using boxes with sides that are 1/2 the size of the previous boxes. The capacity dimension is then the slope of the plot of Log N(x) versus Log (1/r). or equivalently, the negative of the slope of the plot of Log N(®) versus Log (x). If an object is self-similar, then the slope of Log N(r) versus Log (1/r) is the same as the limitof Log N(x) / Log (1/1) as approaches 0. It is much easier to determine the slope than the limit. New algorithms make it possible to determine efficiently the number of boxes that contain at least some part of the object. Using these new algorithms, box counting is a particularly good method to evaluate the fractal dimension of images in photographs. 56 Fractals and Chaos Simplified for the Life Sciences 1.46 Fractal Dimension of an Object by Box Counting r = Box Size N(1) = Number of Boxes Needed to Cover the Set r= 1/2 N=3 1/4 N=11 118 N=26 FEC N(1) = 1.03 r 1-60 a t q ALog N(r) “ ALog (1/r) __ ae = 1.60 A Log (r) Fast Box Counting Algorithms: Liebovitch & Téth 1989 Phys. Lett. A141:386-390. Hou et al. 1990 Phys. Lett. A151:43-46. Block et al. 1990 Phys. Rev. A42:1869-1874. ST147 FRACTALS / Dimension Fractals and Chaos Simplified for the Life Sciences 147 Example of Determining the Fractal Dimension: Fractal Dime nsion Dete rm i ned Using the Scaling Relationship from the Scaling Relationship ‘The fractal dimension can be found from the sealing relationship. This is the most often used method to determine the fractal dimension from in general: experimental data. ‘The fractal dimension d mcans that the number of pieces N(r) measured N(r) xf -d dimension: N(r) is the number of pieces at resolution r is proportional to r4, The scaling relationship evaluated found at resolution r from the experimental data tells us that the value of a property Q(t) measured at resolution r is proportional to >, If we know how the property Q{r) . depends on the number of pieces and the size of each piece, we can determine Qin «fT b scaling relationship of the property Q(1) the fractal dimension d from the exponent b in the scaling relationship. measured from the data For example, Richardson measured the length of the coastline of Britain by Qa layingline segments of length r end to end along the coastline. The length r Qin oc [N (f)] [rf] B theory von how propery 29) of the line segments set the resolution used to make the measurement. The total length of the coastline Q{r) is equal to the number N{r) of these line segments multiplied by the length r of each one. That is, Q(r) = rNt). The definition of the fractal dimension is that the number of line segments Nir) is Thus, the dimension: proportional to r@, Thus the total length of the coastline Q(r) is proportional to rid. Richardson repeated the measurement using line segments of different size r. In this way he determined the scaling relationship that the total length of the coastline Q(t) was proportional to 1-25, Equating the exponent of the scaling relationship to the exponent determined from the properties of the dimension, we find -.25=1-d. Thus the fractal dimension for the length of the coastline of Britain: of the length of the coastline is equal to 1.25. N(t) —ooo— a-2 6114.9 FRACTALS / Dimension The Embedding Dimension ‘The embedding dimension describes the fractal dimension of the space that contains the fractal object. Usually, fractal objects are present in spaces whose dimensions are 1, 2. or 3. For example, the points in time when an fon channel switches open or closed are embedded along a 1-dimensional time line. The retina is very thin. ‘Thus, the nerves and the blood vessels in the retina are embedded in a 2- dimensional space. The tubes that bring air into the lung are spread out into 3-dimensional space. However, fractal objects can also be present inside other fractal objects. For example, there are fractal patterns of the motion of molecules along the fractal surface of a solid catalyst. 62 Fractals and Chaos Simplified for the Life Sciences 1.4.9 Embedding Dimension Fractals can live inside integer dimension spaces: A chemical reaction can occur in 1, 2, or 3 dimensional space. Fractals can live inside non-integer dimension spaces: A chemical reaction can also occur in a fractal dimensional space. 631.4.10 FRACTALS / Dimension Fractals and Chaos Simplified tor the Life Sciences 1.4.10 Definition of a Fractal Definition of a Fractal A fractal is an object in space or a process in time that has a fractal dimension greater than its topological dimension. d (fractal) > d (topological) For example, the perimeter of the Koch curve has a fractal dimension of about 1.2619. The fractal dimension describes the space filling properties of the perimeter. The fractal dimension of the perimeter is between 1 and 2. Since the fractal dimension is larger than 1, the perimeter is so wiggly that it covers more than just a I-dimensional line. But since the fractal dimension is less than 2, the perimeter is not so wiggly that it covers a 2-dimensional area, example: perimeter: d (fractal) 1.2619... d (topological) = 1. ‘The topological dimension of the perimeter is 1. The topological dimension describes how the points on the perimeter are connected together. No matter how wiggly the perimeter is, it is still a line with a topelogical dimension equal tol Since the fractal dimension of the perimeter (1.2619) is greater than the topological dimension of the perimeter (1), the perimeter of the Koch curve is a fractal. 1.2619... > 1. d (fractal) > d (topological) ‘The topological dimension tells us what kind of thing an object is, such as an edge, a surface, or a volume. When the fractal dimension is larger than the topological dimension, it means that the edge, surface, or volume has more finer pieces than we would have expected of an object with its topological dimension. It is more wiggly than we expected. That is why we keep seeing more smaller pieces when we examine it at finer resolution. perimeter: covers more space than a 1-D line “The additional emaller pieces at finer resolution mean that the object covers more space than we would have expected of an object with its topological covers less space than a 2-D area dimension. The topological dimension is an integer. The additional space covered means that the fractal dimension is an integer plus an additional fraction. Mandelbrot says that he coined the word “fractal” to reflect these central ideas that a fractal is: (1) fragmented into ever finer pieces and (2) has a "FRACTAL" fragmented, many pieces fractional dimension. fractional dimension 64 as1.4.11 FRACTALS / Dimension Example of the Fractal Dimension: Blood Vessels in the Lungs Many parts of the body have fractal structures. The fractal dimension can be used to measure the differences between normal structures and those altered by disease. Laboratory rats that are brought up breathing less oxygen or more oxygen: than is found in normal air have higher blood pressure in the blood vessels in the lungs. X-ray images can be taken of these blood vessels. We digitized these images and used the box counting method to determine the fractal dimension of these blood vessels. We found that the fractal dimension of the blood vessels in the normal lungs was 1.65. We also found that the fractal dimension of the blood vessels in the abnormal lungs was 1.53 and 1.43, That is, the fractal dimension was lower in the abnormal lungs. ‘The fractal dimension tells us how many additional branches are found as smaller blood vessels are examined. Thus there were fewer, finer branches in the blood vessels from the abnormal lungs. ‘The fractal dimension is one way to measure the differences between these normal and abnormal lungs. 66 Fractals and Chaos Simplified for the Life Sciences 14.11 Pulmonary Hypertension HIGH BLOOD PRESSURE IN THE LUNGS Boxt, Katz, Czegledy, Liebovitch, Jones, Esser & Rei Katz, é : ‘eid 1994 J. Thoracic Imaging 9:8-13 normal D=1.65 20% Op hypoxic D=153 10% O> hyperoxic D=1.43] penx 90% Oy 671.4.12 FRACTALS / Dimension More Examples of the Fractal Dimension We have already seen that the fractal dimension can be used to measure the branching pattern of the blood vessels in the lung. Fractal dimensions have also been measured for many structures and processes at the molecular, cellular, and organ levels. ‘The most commonly used methods for determining these fractal dimensions have been the use of the scaling relationship and the capacity dimension done by box counting. Examples where the fractal dimension has been measured include: 1. The surfaces of proteins, cell membranes, cells of the cornea damaged by infection with the herpes simplex virus, and bacterial colonies; 2. The branching patterns of one type of lipid within the rest of the lipids in the cell membrane, the arms (dendrites) of nerve cells, and the blood vessels in the eye, heart, and lung; 8. The distribution in space of the blood flow in the heart, the density of bone and teeth, and the concentration of radioactive tracer in the liver: 4. The distribution in time of the intervals between the electrical impulses in nerve cells and the opening and closing of ion channels; 5. The distribution in energy space of the differences in energies of vibration in proteins; and 6. The concentration dependence of the rates of chemical reactions of protein enzymes. 68 Fractals and Chaos Simplified for the Life Sciences 14.12 Biological Examples where the Fractal Dimension has been Measured surfaces of proteins surface of cell membranes shape of herpes simplex ulcers in the cornea growth of bacterial colonies islands of types of lipids in cell membranes dendrites of neurons blood vessels in the eye, heart, and lung blood flow in the heart textures of X-rays of bone and teeth texture of radioisotope tracer in the liver action potentials from nerve fibers opening and closing of ion channels vibrations in proteins concentration dependence of reaction rates of enzymes 691.4.13 FRACTALS / Dimension Biological Implications of the Fractal Dimension 4. Quantitative Measure of Self-Similarity ‘The numerical value of the fractal dimension d_ gives us a quantitative measure of self-similarity. The fractal dimension tells us how many small pieces Nir) are revealed when an object is viewed at finer resolution x. The quantitative relationship is that N(z) is proportional to rd. The larger the fractal dimension, the larger the number of small pieces are revealed as the object is viewed at finer resolution. 2. Quantitative Measure of Correlations “The small pieces of a fractal are each related to the large pieces. Thus the small pieces are also related to each other. The fractal dimension measurss the correlations between the small and large pieces. Thus it also measures the correlations between the small pieces themselves. 3. Quantitative Classification ‘The numerical value of the fractal dimension can be used to classify different objects. For example, the fractal dimension of the blood vessels in a hhormal retina is different from the fractal dimension in a retina changed by disease. The fractal dimension may serve as a method to diagnose different diseases and as an index to quantify the severity of these diseases. 4. Hints About Mechanisms Different mechanisms produce fractals with different dimensions. Hence, measuring the fractal dimension of an object may give us clues about figuring Gut the mechanism that produced it. For example, the process called diffuston Himited aggregation produces fractals with a fractal dimension of about 1.7. ‘The blood vessels in the retina have a fractal dimension of about 1.7. Thus it is worthwhile to consider if the growth of these blood vessels was produced by diffusion limited aggregation, This would mean that the growth of the blood Vessels was proportional to the gradient of a diffusible substance, such as oxygen or a growth factor. 70 Fractals and Chaos Simplified for the Life Sciences 1.4.13 Fractal Dimension Numerical Measure of Self-Similarity Numerical Measure of Correlations in Space or Time ALL Numerical Measure of Normal versus Sick D=17 Mechanism Mechanism: Diffusion Limited Aggregation 711.5.1 FRACTALS / Statistical Properties The Statistical Properties of Fractals Fractals may seem to have strange statistical properties. These properties really aren't strange. ‘They only appear to be strange because they are not cee a in the usual statistics courses. Those courses only covet the properties of statistical distributions that are called Gaussian or asymptotically Gaussian distributions. The statistical properties of fractals Delong to a more general class of distributions which are called stable Gistribotions. Although few scientists scem to be aware of the propertics of aie distributions, mathematicians have studied them for over 250 year. Tor exomple, the average of a fractal may not exist. How can the average not exist? If we have 5 measurements, can’t we just punch those 5 values into w hand calculator, determine their sum, divide the sum by 5, and get the average? “That calculation on the hand calculator alone is not enough to demonstrate that the average exists. The average from one set of data that we determine saan dur hand calculator is called the sample mean. ‘To show the average Wiets we must show that as more data are analyzed, these sample T=. Seach @ Limiting value. We then consider that limiting value a5 ‘the “real” average of the thing that we measured, This “real” average is called the population mean. 4. Non-Fractal For non fractal objects, as more data Is included, the averages of the data reach a limiting value which we therefore consider to be the “real” averati {That is, the sample means converge to a finite, nonzero, Hmiting value that we identify as the population mean.) 2. Fractal However, for a fractal, as more data are included, the averages of (he data will continue to increase or continue to decrease. Thus there is no limiting Witue, we ean consider to be the “real” average. ‘The average does not exist. (That is, the sample means do not converge to a Gite, nowero- limiting vaise, and thus there is no value that we can identily as the population mean.) 74 Fractals and Chaos Simplified for the Life Sciences p90 Non-Fractal Mean hoi More Data > Fractal Mean Mean [0 More Data -> More Data —-» 7515.2 FRACTALS / Statistical Properties Self-Similarity Implies That the Moments Do Not Exist Selfsimilarity means that there are an ever larger number of smaller pieces that resemble the Jarger pieces. Thus there are a few big pleces, many medium sized pieces, and a huge number of small pieces. When we determine the average of the values of the sizes of the pieces, the most significant contribution to the average will come from either: (1) the frequent number of small values, in which case the average decreasis aS Mere data are included, or (2) the infrequent large values, in which case the average increases as more data are included. ‘A moment is the average of a property raised to a power. ‘The mean is the first moment. Itis the average of the values of the sizes of the pieces. Self-simlarity means that the mean will increase or decreas’ 36 more data are analyzed, depending on the relative contribution of the frequent small values versus the infrequent large values. ‘The varlanee is the second moment. It is the average of the square of the difference between each value and the mean value of the sizes of the pieces. Typically, self-similarity means that the variance will increase as more data are analyzed because there are an ever larger number of self-similar fluctuations. We are so used to the properties of Gaussian distributions, where the mean and variance exist, that we assume that these moments must always exist. ‘The exstence of the mean and variance depend on the data satisfying the mathematical assumptions that these moments approach finite, nonzero Hmniting values as more data is analyzed. This fact 1s not well known. When the data are fractal, it does not satisfy these assumptions. the mean and Varianee do not exist, and therefore they are not useful to characterize the properties of fractal data. 76 Fractals and Chaos Simplified for the Life Sciences 15.2 The Average Depends on the Amount of Data Analyzed ‘each piece pi e 4 ese @ average Ee average or| size number of pieces included number of pieces included ne fix1_+ fexe + fox + Ta een { is the number of pieces of size x tim jim = ix1_+ fexe + fp mean = Aix_+ (92 + fxg +. Hin x0 x0 ++ +...4 fn = 0 Contributions to the mean dominated by the many f, of the smallest sizes x1. ste mean = lim fix1_+ foxe + fxg +... +fikn Se fi+beh+...+ “8 Contributions to the mean dominated by the few f, of the biggest sizes xp. 7715.3 FRACTALS / Statistical Properties Example Where the Average Does Not Exist: The St. Petersburg Game 1, The Average Winnings is 50¢ fora Non-Fractal, Ordinary Coin Toss Game Play the following game. Toss a coin. If it lands Heads, you win $1; if it lands Tails, you win nothing. The average payback to you is the sum of the probability of each outcome multiplied by the winnings associated with it, namely (1/2) x 0 + (1/2) x 1 = $0.50. The more times N that you play this game, the closer the payback to you, averaged over all the N games, approaches $0.50. The house should be ready to pay out $0.50, on average, and you should be willing to bet $0.50 to play each game. 2. There is No Average Winnings for the Fractal, St. Petersburg Game Now play a game formulated about 250 years ago by Niklaus Bernoulli and analyzed and published by his cousin Daniel Bernoulli. You toss a coin and continue to do so until it lands Heads. You get $2 from the house if it lands Heads on the first toss, $4 if on the second toss, $8 if on the third toss, $16, if on the fourth toss, and so on, so that with cach additional toss the number of dollars that the honse must pay is doubled. The average payback to you is (1/2) x2 + (1/4) x 4+ (1/8) x 8 + 0/10) X16... = 1H LEL ELA rm There is NO number that we can identify as the average Winnings! ‘The more times N_ that you play this game, the larger the payback to you averaged over all the N games. The average winnings after N games continues to increase as N increases. It does not reach a finite, imiting value that we can identify as the average winnings. The average winnings for playing this game does not exist. ‘This game is fractal because the distribution of winnings (how much you win for each probability of winning) has a power law scaling ‘This game is called the St. Petersburg paradox. Since there is a 50/50 chance that you will $2 on each game, you would be willing to put up $2 to play. But since the average winnings is infinite, the house would want you to put up more than all the money in the world to play. This type of game and the statistical properties associated with it were studied by mathematicians since it was first proposed. However, these ideas became separated from the main thread of probability theory that became popular among natural scientists. 78 Fractals and Chaos Simplified for the Life Sciences 15.3 Non-Fractal Ordinary Coin Toss Toss a coin. If it is tails win $0, if itis heads win $1. The averagewinnings are: 91, - 5 Toss a coin. If itis heads win $2, The average winnings are: 2°'2' 4 2222 + 9953 , St. Petersburg Game (Daniel Bemoulli) Feller 1968 An Introduction to Probability Th Applications, vol. 1, Wiley, pp. 251-3. Bahecyanalte if not, keep tossing it until it falls heads. If this occurs on the N-th toss we win ¢ 2 H 82 ai TH $4 With probability 2°. we win $2 N. ie TITH $16 etc. To# 4 Ft 4 4. co 100 ~ St. Petersburg Game Mean Fractal Winni NINgS 4, after N Trials 1 “tho, Non-Fractal fens ee 5 BORO O-O HOCH} a anand Ordinary Coin Toss ° to” 40° 40° 40° i0* 10° 10° 407 N Trials 7915.4 FRACTALS / Statistical Properties Example Where the Average Does Not Exist: Diffusion Limited Aggregation (DLA) 1. The Average Density of a Non-Fractal Checkerboard We want to determine the average density of the pixels in a black and white checkerboard. To find the average density within a circle of radius 1 we count the number of black pixels within the circle and divide that by the total number of pixels. We repeat this measurement for circles of larger radii. As the radius increases, we find that there are some fluctuations in these averages. However, we also find that as the radius increases, these average Gensities approach a finite, limiting value that we therefore identify as the “real” average density of the checkerboard. 2. There is No Average Density of a Fractal DLA ‘We now want to determine the average density of a fractal object called a diffusion limited aggregate (DLA). It is formed by particles that are released one at a time from far away and randomly walk until they hit and stick to the growing structure. The DLA is self-similar. It has little spaces between the small arms and bigger spaces between the larger arms. We measure the average density within a circle of radius 1 As the radius of the circle increases, we catch more of the ever bigger spaces between the ever larger arms. Thus, as the radius r of the circle increases, the average density within radius r decreases. The average density docs not reach a finite, limiting value. It approaches zero. Therefore, the “real” average density for the DLA does not exist. 80 Fractals and Chaos Simplified for the Life Sciences Non-Fractal 1.5.4 Log avg density within radius r Log radius r Fractal Meakin 1986 In On Growth and Form: Fractal and Non-Fractal Patte ir ic Ed, Stanley & Ostrowsky, Marinus Nijoff Pub. p18? Log av, i density ° within radius r Log radius r a115.5 FRACTALS / Statistical Properties Example Where the Variance Does Not Exist: The Roughness of Rocks The edges of rocks are self-similar. Small sharp peaks over short distances are reproduced as ever larger peaks over ever larger distances. The root mean square (RMS) of the heights along the rock profile is the square root of the average of the square of the heights along the rock. As the root mean square of the heights is measured over longer distances on the rock, ever larger peaks are included. Thus the root mean square of the heights increases with the distance measured along the rock. It does not approach a finite, limiting value. ‘The root mean square (RMS) of the heights along the rock is related to the variance of the fluctuations in height along the edge of the rock. For a non-fractal object, as the variance is measured over longer distances, it approaches a finite value. Thus the variance exists and we identify that. limiting value as the “real” value of the variance. However, for the fractal edges of rocks the variance increases with the distance measured. The variance does not approach a finite value. Thus the variance does not exist. 82 Fractals and Chaos Simplified for the Life Sciences 15.5 Fractal Edges of Rocks Brown and Scholz 1985 J. Geophys. Res. 90:12,575-12,582. Height Distance tol 108 Profile Length (u) 831.5.6 FRACTALS / Statistical Properties Example Where the Variance Does Not Exist: Electrical Activity in Nerve Cells ‘Transient changes in the electrical voltage across the cell membrane, called action potentials, travel along the surface of nerve cells. Information is passed down the nerve axon in the timing between the action potentials. ‘Teich et al. analyzed the timing of the action potentials from nerves in the car connected to the auditory cells that encode information about sound and from nerves connected to the vestibular cells that encode information about acceleration that is used for balance. They divided their records into consecutive time windows and counted the number of action potentials in each window. They determined how the number of action potentials in each time window depended on the length of the time windows. The tme resolution was set by the length of the time windows. ‘Thus they determined how the properties of the timing of the action potentials depended on the time resolution used to measure them. For example, they studied the number of action potentials per second in each time window, which is called the firing rate. They evaluated the firing rate for time windows of different lengths. For a non-fractal, the fluctuations in the firing rate would be less in longer windows, because the statistical inaccuracies would average out. Instead, for the auditory nerve cells involved in hearing, they found that the variation in the firing rate decreased more slowly than for a non-fractal as the window length increased. ‘This result is surprising if you only know about the statistical properties of non-fractal distributions. For those distributions the fluctuations decrease at a certain rate as more data are included. However, for fractal distributions, the fluctuations decrease more slowly as: more data are included. Thus their result implies that the firing rate is fractal. ‘This means that the action potentials are correlated with cach other. These small correlations become more significant when the action potentials are analyzed over longer times in the longer time windows. B84 Fractals and Chaos Simplified for the Life Sciences 1.5.6 Electrical Activity of Auditory Nerve Cells Teich, Johnson, Kumar, and Turcott 1890 Hearing Res. 46:41-52 Data: acti i voltage ction potentials time Divide the record into time windows: Count the number of action potentials in each window: LL WET Td ie 6 3 Firing Rate = 2, 6,3, 1, 5,1 7 Repeat for different lengths of time windows: LP MU et Firing Rate = 8, 4,6 The variation in the firing rate decreases slowly at longer time windows. oo 4 8 2 tm ee SAMPLE HUMBER (each of duration T sec) 85157 FRACTALS / Statistical Properties Statistical Analysis of the Electrical Activity in Nerve Cells 1. Pulse Number Distribution Teich et al. divided their records of action potentials recorded from nerves in the ear into consecutive time windows. They counted the number of action potentials in cach time window. They evaluated how often each number of action potentials occurred in these time windows. This is called the pulse number distribution. For a non-fractal process, as the window size is increased. the pulse number distribution becomes smoother and more like a Gaussian distribution. This property is called the Central Limit Theorem. This was the case for the data from the vestibular cells involved in balance. This was not the case for the data from the auditory cells involved in hearing, Those distributions became rougher as the length of the time window was increased. This roughness ariscs from correlations in the times between. the action potentials. ‘The proof of the Central Limit Theorem requires that the variance exist and that it has a finite value. For a fractal, the variance is not defined, and thus the Central Limit Theorem does not apply. As more data are included fractal distributions do not become smoother or more Gaussian. Rather, the distributions become rougher because the correlations that link the deviations together in a self-similar way become more noticeable over longer times. Teich et al. wrote that “the irregular nature of the long count pulse number distributions does not arise irom statistical inaccuracies associated with insufficient data, but rather from event clustering inherent in the auditory neural spike train.” 2. Fano Factor ‘The Fano factor is equal to the variance divided by the mean of the number of action potentials in the time windows. For a non-fractal process where the time between the action potentials is random, the pulse number distribution is a Poisson distribution, and the Fano factor is 1. They found that the Fano factor increased with the length T of the time windows used to measure it. ‘The Fano factor F had a power law scaling relationship that F was proportional to T4 where d is the scaling exponent. 86 Fractals and Chaos Simplified for the Life Sciences 15.7 Statistical Analysis of Acti i i cti Teich 1989 IEEE Hed Biomed. reton Potentials LTT : 2 4 3 1 | | A | Counts in the windows are: 2, 4,3,1, 1,4 3. Determine the Mean and 2 Variance of the counts. 7 Determine the Pulse Number Distribution: Number (i) of windows of size T with n counts, ° o 123 4 1 counts in a window Pulse Number Distribution \Non-Fractal| as,T T the PND . T=50ms = Vestibular | becomes more Teno Nears Gaussian Fractal asTT thepND 3 i of ; T= 50m: = Auditory becomes loss 7 i. reno me Neuron coe Fano Factor = Variance/Mean a0 = weneman $ 5 Techand g : kerala The Fano Factor = 3 Variance/Mean gi Te WT increases with the i window length T for 2 the auditory igh a neurons. Wo wt 10 ao ae Window length, 7, seconds 8715.8 FRACTALS / Statistical Properties Example Where the Variance Does Not Exist: Blood Flow in the Heart Bassingthwaighte et al. measured blood flow in the heart by adding radicatlive travers to the blood. These tracers leak out of the blood vessele and become trapped in the muscle of the heart. The amount of tracer that leaks out is proportional to the blood flow in that region. ‘Thus they determined the blood flow by measuring the amount of radioactivity in different pieces of the heart. ‘They determined the relative dispersion, RD, which i equal to the standard deviation divided by the average of the blood flow measured in & piece of weight w. They measured the blood flow in pieces of different weight. The weight of the pieces set the spatial resolution of the vresguement, They found that the relative dispersion had the power law scaling relationship that RD was proportional to wid, where d is the fractal dimension. ‘Thus blood does not flow evenly throughout the heart. The flow of blood through the heart is fractal. ‘There are regions of higher than average blood flow and regions of lower than average blood flow. ‘The pattern of blood flow is seltsimilar There are ever smaller regions of higher and lower than average blood flow. The dimension d is a measure of the correlation between regions with different amounts of blood flow. ‘They showed that if the large blood vessels connect to slightly unequal smalles branches, then the relative dispersion would increase after cach branching. This would explain why the relative dispersion increases as it is preasured from pieces of smaller weight that correspond to the finer spatial resolution. Fractals and Chaos Simplified for the Life Sciences 15.8 Blood Flow in the Heart Bassingthwaighte and van Beek 1988 Proc. IEEE 76:693-699 When the blood flow is measured using smaller pieces, the relative dispersion increases. Data “ ee re RD = Standard deviation g mean & 2 RD « mt-d | d=12 syste of sue tamples.grams Model 5+e (1-ay(r-ayr 891.5.9 FRACTALS / Statistical Properties Example Where the Variance Does Not Exist: Volume of Breaths Hoop et al. measured the volume of alr in cach breath of a rat We used the seveeled range analysis developed by Hurst to analyze the values of the aon ree of conserutive breaths. The rescaled range R/S $sa measure of the seamce, We evaluated the rescaled range over time windows of different Tength ‘The length T of the time windows is called the lag ‘The values of the lag set the temporal resolution. Thus we determined how the rescaled Tenge depended on the temporal resolution at which s{ was measured |A fractal set of values has small variations in the values over brief ttsnes thot look similar to larger variations in the values over longer times. All the past values affect the future values. The rescaled range R/S of a fractal time ea tee will have che power law scaling relationship that R/S is proportional to TH, where H is called the Hurst exponent. When H = 1/2, the differences between consecutive values are unegreelated, When 1/2 | killer vi +] Killer virus | How much is the variation in the number of mutant cells? BIG variation (Lea & Coulson Distribution) urTLe variation (Poisson Distribution) 9315.11 FRACTALS / Statistical Properties The Distribution of Mutations Is the Same as in the St. Petersburg Game If we run many experiments where mutations occur all the time, what is the average number of mutant cells at the end of the experiments? Each cell has the same probability of giving birth to a mutant cell. There are few cells at the beginning of the experiment. Thus there is a low probability that a mutation will occur in the first generation of cells. However, if a mutation does occur in the first generation, that cell will produce 2" daughter cells in the N_ generations until the end of the experiment. In the second generation, there are already twice as many cells as the first generation. Thus the probability that a mutation occurs in the second generation is twice as great as that in the first generation. However, that mutant cell will produce only 2N-1 daughters in the remaining N-1 generations. Similarly, for all subsequent generations. ‘The total number of mutant cells at the end of the experiment is equal to the probability of a mutation occurring in each generation multiplied by the number of daughter cells that it produces until the end of the experiment. ‘This means, on average, that each generation contributes the same number of mutant cells to the final number of mutant cells. This is the same calculation as determining the winnings in the St, Petersburg game described previously. Each generation corresponds to one play of the game, the probability of winning on that play corresponds to the probability that a mutation occurs in ‘that generation, and the money won corresponds to the number of resistant daughter cells af the end of the experiment, As is true for the average winnings of the St. Petersburg game, the average number of mutant cells at the end of the experiment is not defined. The average found for a number of experiments will increase as the number of experiments is increased. Because of this large variability, it is not known how to compare the average number of final mutant cells from two sets of experiments to determine if the different conditions of the experiments affected the mutation rate. 94 etc. Fractals and Chaos Simplified tor the Life Sciences 1.5.11 Random Mutations Lea & Coulson 1949 J, Genetics 49:264-85, Mandelbrot 1974 J, Appl. Prob. 11:437-444, If the probability of a mutation Per cell 1 tei ~ 16° 2 FOR EACH GENERATION: Number of Mutants at the end from this generation | 2 a | 1 i 23 1 | e ae 4 1 | i 2 gl 1 / 2° 1 Prob, wits Total St. Petersburg Game co 951.6.12 FRACTALS / Statistical Properties Statistical Properties and the Power Spectra A sequence of values measured in time can be broken down into scrics of periodic oscillations of different frequencies. A plot of the amount present in each frequency is called the power spectrum. ‘The power spectra of fractal objects in space or fractal processes in time reflect the self-similarity, scaling relationship, and statistical properties of fractals. Low frequencies in the power spectrum correspond to coarse resolution. High frequencies in the power spectrum correspond to fine resolution. Thus the power spectrum is a measure of the amount present in structures of different sizes, Self-similarity means that there is a relationship between the power at high frequencies (fine resolution) and the power at low frequencies (coarse resolution). ‘This relationship has a power law scaling that the energy at a given frequency is proportionalto 1/f%. This form is called 1/f (‘one-over-f) noise, even when the exponent o is not equal to 1. Depending on the value of a, the total power at all the frequencies in the power spectrum and the average power de not exist. When «> 1, then the total energy increases as the lowest frequency used to measure it decreases. ‘That is, the longer the interval of data that is analyzed, the larger the total power in the power spectrum. When « < 1, then the total power increases as the highest frequency used to measure it increases. In this case, the ever shorter intervals of data contain ever larger amounts of power. An example of 1/f noise in time is the electrical signal generated by the contraction of the heart. An example of 1/f noise in space is the spatial distribution of a radioactive tracer in the liver. 96 Fractals and Chaos Simplified for the Life Sciences 15.12 Fractal Power Spectra P(f) If the variance - ©°, then there are fluctuations at ALL scales, and P(f) = in time electrical activity when the heart contracts Goldberger, Bhargava, West, and Mandell 1985 Biophys. J. 48:525-528 in space (Rs sPECrAUH, Ca a ee Aogtharmanie) radioactive isotope distribution in the liver Cargill, Barrett, Fiete, Ker, Patton, and. ‘Seeley 1988 SPIE 914 Medical Imaging II, pp. 355-361, Slope=-3.95 Diagnosis: NORMAL & LoarPowen spectrum) ent or os " LoatsraTiaL FREQUENCY) 9718.13 FRACTALS / Statistical Properties How to Measure the Properties of Fractal Data ctal The statistical properties. such a the average Seas oe aaa "i object or process depend on the resolution use doce Mette good to measure these statistical properties at only one resolution. ‘The meaningful measurement is to determine how these statistical Properties depend on the resolution used to measure them. Usually, the scaling relationship for a statistical property Q{r) will have the Jaw form that Q(1) is proportional to 1°, where ris the resolution. power law form tha The exponent b is related to the fractal dimension. Anumber of different statistical properties Q(r) have been used to analyze fractal data, including the: 1, mean (average): 2. variance; 8. standard deviation; 4. relative dispersion (standard deviation /variance); 5. Fano factor (variance/mean); 6, mean squared deviation; and 7. rescaled range (range of the running sum of the deviations from the mean divided by the standard deviation). 98 Fractals and Chaos Simplified for the Life Sciences 15.13 When the moments, such as the mean and variance, don't exist, what should I measure? You should measure how a Measure this slope Property Q(r) Log Q(r) depends on the resolution r used to measure it. examples of Q(r): Logr Mean €.g. average density within a circle as a fun iction of the radius of that circle Meakin 1986 In On Growth and Form, ed. Stanley & Ostrowksy Nijhoff. pp. 111-135 Relative Dispersion (standard deviation / mean) &9. relative dispersion of blood flow as a function of the mass of the tissue sample Bassingthwaighte and van Beek 1988. Proc. IEEE 76:693-699 Fano Factor (variance / mean) .g. Fano factor of the number of action the length of time windows Teich, Johnson, Kumar, and Turcott 1990 Hearing Res. 46:41-52 Potentials within time windows as a function of Mean Squared Deviation °-9: mean squared deviation of a walk generated from the base pair sequence in DNA as a function of the length along the DNA. Peng etal. 1992 Nature 356:168-170. Hurst Rescaled Range 9. maximum minus the minimum value of the running sum of tho deviations from the Stine normalized by the standard deviation of volumes of breaths measured within a time window as a function of the length of th it Hoop et al. 1993 Chaos 3:27.29, 99Fractals and Chaos Simplified for tho Life Sciences 1.5.14 1.5.14 FRACTALS / Statistical Properties More Examples of the Statistical Properties of Fractals Biological Examples of the Meher area seen dat te stteiot pein otiattae meses Statistical Properties of Fractals in the timing of the action potentials recorded from nerves that cary information about sound that we use for hearing, in the spatial distribution of blood flow in the muscle of the heart, in the sequence of the volumes of breaths, in the number of mutated cells produced by Darwinian evolution, in the timing of the electrical activity of the heartbeat, and in the spatial distribution of radioactive tracer in the liver. Additional examples of these statistical properties include the changing action potentials in nerve cells electrical voltage across the cell membrane of white blood cells (T-lymphocytes), the sequence of base pairs in DNA where cach base pair is blood flow in the heart assigned a number and the running sum of these numbers is analyzed, and * the duration in time of consecutive breaths. volumes of consecutive breaths mutations electrical activity of the heartbeat distribution of tracer in the liver membrane voltage of T-lymphocytes base pair sequence in DNA durations of consecutive breaths 101 1001.5.15 FRACTALS / Statistical Properties Biological Implications of the Statistical Properties of Fractals 1. Not Gaussian or Asymptotically Gaussian ‘The statistical knowledge of most scientists is limited to the statistical properties of Gaussian distributions. Fractals do not have the properties of Gaussian distributions. In order to understand the many fractal objects and processes in the natural world, it is required to learn about the properties of stable distributions. Stable distributions are more general than Gaussian distributions. 2. The Average and Variance Do Not Exist ‘The moments of a fractal, such as the mean and variance, do not exist. AS more data are included, the measurements of these moments do not approach finite, limiting values. 3. Large Variations ‘The variance of a fractal increases as more dataare analyzed. The average yalues measured for the properties of the data will have wide variation from one time to another and among repetitions of the same experiment, 4, When Are These Large Variations Significant? When the variance of a fractal increases as more data are analyzed, we do not know how to perform statistical tests to determine if the parameters of the mechanism that generated the data have changed from one time to another or between experiments run under different conditions. ‘The statistical tests taught in the usual statistics courses are based on the assumption that the variance is finite. These tests are not valid to analyze fractal data where the variance is infinite. It would be very worthwhile for mathematicians to formulate statistical tests for fractal distributions where the variance is infinite. 102 Fractals and Chaos Simplified for the Life Sciences 1.5.15 Statistical Properties What they taught you in school moments — nonzero, finite o> finite Statistical tests to tell if parameters differ at different times or between different experimental conditions: 1, F, ANOVA non-parametric 103 Fractals Stable Distributions moments — 0, statistical tests to tell if _ parameters differ at different times or between different experimental conditions: ?1.5.16 FRACTALS / Statistical Properties Biological Implications of the Statistical Properties of Fractals (continued) 5. “Nonstationary” Does Not Mean That Things Are Changing When the measurements of the moments keep changing as more data are included, then a process is said to be “nonstationary.” ‘This is a poot choice ofa word, because it suggests that the process is changing in time. “This is pot necessarily true, A generating mechanism whose operation is fixed in Time can produce « fractal output whose moments keep changing in time When the moments of the data are “nonstationary.” this does not necessarily mean that the mechanism that produced the data (s changing in time. 6. When the Variance Looks Big, Maybe It’s Infinite Sometimes there is a large variability in a property measured in repetitions of an experiment, For example, such variability is present in the nurnber of orutant eells in evolution and in the time durations between the steps of growth sand divicion (cell cycle time) in cancer cells. Selentists analyzing this Kind of aoes who are familiar only with Gaussian statistics usually assume that the Gata were produced by a process with a finite variance. Any me such dats ‘with large variability are found, it may be worthwhile to determine if the valiant dees, or does not, have a Gnite, limiting value. This can be done by Tneasuringhow the variance depends on the amount of data included. 11 the vaniance increases with the amount of data included, then the data have fractal propcrtics and the variance does not exist. 7. importance of the Scaling Relationship When the moments, such as the mean and variance, do not exist, their values will depend on the resolution used to measure them. Thus the vneasurement of the moments at one resolution fs not meaningful. What is Ineaningfulis to determine bow the moments depend on the resolution used to measure them, The form of this dependency is called the scaling Foutionship, and it is characterized by the parameter called the fractal dimension. 104 Fractals and Chaos Simplified for the Lite Sciences 1.5.16 More Statistical Lessons nonstationary ; mean This word means that the moments do not exist. (The moments do not reach finite, limiting values.) Itdoes not mean that the mechanism th at produced the data is changing in time. time BIG variance? Check its limiting value. Maybe, it's -. Bad: one measurement Good: slope E = S @ ~<— one value = slope. 5 5 E E = = 3 & = 3 Logarithm of the resolution Logarith: ‘i used to make the measurement usedio make ite moseurenent 105Ba FRACTALS Summary 1.6.1. Summary of Fractals 1.6.2 Where to Leam More about Fractals1.64 FRACTALS / Summary Summary of Fractals 1. Fractals Are Self-Similar Self-similarity means that ever smaller pieces of an object resemble the larger pieces of an object. 2. Self-Similarity Produces a Scaling Relationship When the value of a property is measured at finer resolution, it will include ever more of the smaller, self-similar pieces. Thus the value measured for a property depends on the resolution used to make the measurement. This is called the scaling relationship. The scaling relationship usually has the power law form that the value Q() measured for a property at resolution r is proportional to 3. The Dimension is a Measure of Self-Similarity and Scaling The dimension is a quantitative measure of the self-similarity and the scaling. The dimension tells us how many new pieces are found when an object is examined at finer resolution. There are many different types of dimension. The fractal dimension characterizes the space filling properties of an object. The different types of fractal dimension share the common feature that the number of pieces N(t) found at resolution r is proportionalto 4, where d is the fractal dimension. 4. Fractals Have Surprising Statistical Properties The statistical properties of fractals are described by stable distributions. Stable distributions are more general than the Gaussian distributions they taught you in school. The moments, such as the average and variance, of stable distributions may not have nonzero, finite values. 108 Fractals and Chaos Simplified for the Life Sciences 1.61 Summary of Fractal Properties Self-Similarity Pieces resemble the whole. Scaling The value measured depends on the resolution. Dimension How many new pieces are found as the resolution is increased. Statistical Properties Moments may be zero or infinite. 1091.6.2 FRACTALS / Summary Where to Learn More about Fractals There are many good books on fractals and hundreds of research articles Published in journals each year. Some references, at different levels, that can Jead you further into the mathematical details and the applications of fractals are the following: 1. Mandelbrot The Fractal Geometry of Nature and its earlier edition Fractals: Form, Chance, and Dimension by Mandelbrot made the scientific community aware of fractals. Mandelbrot developed some of the mathematics of fractals and applied them to many different scientific fields. His book is beautiful, passionate, and at a mathematical level that is sometimes frustratingly too simple and too complex at the same time. 2. Fractals in Mathematics ‘The mathematics of fractals 1s based on measure theory and topology. Two books that provide a clear and rigorous introduction to the mathematics of fractals are Fractals Everywhere by Barnsley and Measure, Topology, and Fractal Geometry by Edgar. 3. Fractals in Physics and Chemistry Fractals by Feder and The Fractal Approach to Heterogeneous Chemistry edited by Avnir provide an introduction to fractals at the mathematical level of the calculus and review applications of fractals in physics and chemistry. 4. Fractals in Biomedical Research Fractal Physiology by Bassingthwaighte, Licbovitch, and West provides an introduction to fractal properties both at a qualitative level and at the mathematical level of elementary calculus. It gives detailed descriptions and references to many biomedical applications of fractals. Additional biomedical applications are described in Fractal Geometry in Biological Systems, edited by Jannaccone and Khokha. 110 Fractals and Chaos Simplified for the Life Sciences 1.6.2 Books About Fractals classic B. B. Mandelbrot The Fractal Geometry of Nature 4983 W.H. Freeman mathematics G. A. Edgar Measure, Topology, and Fractal Geometry 1990 Springer-Verlag M. Barnsley Fractals Everywhere 1988 Academic Pross physics & chemistry J. Feder Fractals 1988 D. Avnir The Fractal Approach to Heterogeneous Chemistry 1989 John Wiley & Sons Plenum biomedical J. Bassingthwaighte, L. Liebovitch, & B. West Fractal Physiology 1994 Oxford Univ. Press P. M. lannaccone & M. Khokha Fractal Geometry in Biological Systems 1996 CRC Press a1Noaawne Part Il CHAOS Deterministic systems with output so complex that it mimics random behavior. X(t+At) = 3.95 X(t) [1-X(1)] a x) t Introduction 115 Phase Space 129 Sensitivity to Initial Conditions 155 Bifurcations 175 Analyzing Data 191 Control of Chaos 225 Summary 237244 CHAOS / Introduction ‘wo Sets of Data That Look Alike Each graph on the facing page consists of a sct of data, Each data value is measured at one instant in time. These instants in time are separated by constant intervals in time. The n-th measurement is denoted by x(n). The values measured are shown as dots. Consecutive values have been connected by straight lines. Both sets of data also have the same statistical properties. Thats, both sets of data have the same average, variance, and power spectrum. Each graph does not have the same values of x(n). However, the pattern of variation in each graph looks similar. The values of x{n) seem to vary ina random way with n. Therefore, it seeims reasonable to believe that both sets of data were generated by an inherently random mechanism. 1. Left: Data Set #1 The values x(n) seem to change randomly from one point in time to the next point in time. 2. Right: Data Set #2 The values x(n) seem to change randomly from one point in time to the next point in time. 116 Fractals and Chaos Simplified for the Life Sciences 244 Data 1 Data 2 x(n) x(n) These two sets of data have the same: mean variance power spectrum 7212 CHAOS / Introduction The Difference between Randomness and Chaos The pattern of x{n) in both data sets appears to have been generated by a random mechanism. However. not everything that looks random really is random. One of these data sets was generated by arandom mechanism. However, the other data set was not generated by a random mechanism at all. It was generated by a deterministic mechanism. 1. Left: Data Set #1 Was Generated by a Random Mechanism ‘This data set was generated by a random mechanism. Each new value of the data x(n) was chosen at random. 2. Right: Data Set #2 Was Generated by a Non-Random, Deterministic Mechanism ‘This data set was generated by a deterministic mechanism. Deterministic means that the next value of the data was computed from the previous values. ‘The next value of x{n+1) was computed from the previous value x(n) by using the simple rule that x{(n+1) = 3.95 x(n) [1 - x(n)]. The phenomenon that a deterministic mechanism can generate data that looks as if it were generated by a random mechanism is called “chaos.” The word chaos was chosen to describe the complex output of these deterministic mechanisms. Chaos is a poor choice of a word for this phenomenon because here it means just the opposite of its common usage of “disordered.” Here, chaos means that the output of a deterministic system is so complex that it mimics the output generated by a random mechanism. It does NOT mean that a system is driven by disorder, randomness, or chance. 18 Fractals and Chaos Simplified for the Life Sciences 24.2 Data 1 Data 2 RANDOM random x(n) = RND x(n) CHAOS deterministic x(n+1) = 3.95 x(n) [1-x(n)] x(n) 1924.3 CHAOS / Introduction ASimple Equation Can Produce Complicated Output ‘The phenomenon of chaos is surprising. Let's follow the deterministic equation, step by step, to see how it produces a set of data that looks random: ‘The first value of x at n=1 was chosen to be x{I) = .892. ‘The value x(n+1) at the (n+1)-th step is computed from the previous value x(n) at the n-th step by using the equation that_x(n+1) = 3.95 x(n) [1 - x(n). ‘Thus, to compute the next value of x at n=2, multiply 3.95 by .892 and then muluply that result by (1-.892). This yields (2) = .380. ‘To compute the next value of x at n=3, multiply 3.95 by .380 and then smultiply that result by (1-.380). This yields x(3) = .931. Continue the same process. ‘This deterministic mechanism alone produces a seemingly random sequence of values x(a). That is why the discoverers of this phenomenon called it chaos. 120 Fractals and Chaos Simplified for the Life Sciences 24.3 x(n+1) = 395[x(n)][1-x(n)] x(n) x(1) = 892 X(2) = 3.95[.892][1-.892] = .380 x(3) = 3.95[.380][1-.380] = .931 x(4) = 3.95[.931][1-.931] = .253 x(5) = 3.95[.253][1-.253] = .747 x(6) = 3.95[.747][1-.747] = .746 etc. 1.0. 0.8. OS. 0.4. 0.2. 0.0. 1 2 3 4 5 6 n 121