Logistic Map: Definitions and Properties

• Logistic Map is a discrete-time nonlinear dynamical system defined by $x_{n+1}=r x_{n}(1-x{n})$ characterized by fixed points, periodic cycles, and chaotic dynamics.
• It provides a canonical example not only of chaos generation but also of periodic windows and invariant measures in nonlinear dynamics.
• Applications and insights from the logistic map include references for physical visualization, cryptography, reservoir computing, and gain–loss systems as a model of self-limited growth.

The logistic map is a discrete-time nonlinear dynamical system defined, in its standard form, by

$x_{n+1}=r x_n(1-x_n),$

where $x_n\in[0,1]$ is the state at iteration $n$ and $r$ is the control, growth, or bifurcation parameter. The map models self-limited growth and provides a canonical example of how a one-dimensional quadratic recurrence generates fixed points, periodic cycles, bifurcation cascades, deterministic chaos, and intricate invariant measures. Its simplicity also makes it a reference model for nonlinear dynamics, symbolic computation, electronic experimentation, stochastic processes, fractional memory systems, reservoir computing, coupled networks, random matrices, and gain–loss instabilities.

1. Definition, fixed points, and primary dynamical regimes

The interval $[0,1]$ is forward invariant for $0\leq r\leq4$. The fixed-point equation,

$x=r x(1-x),$

has the solutions

$x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$

The derivative is

$f_r'(x)=r(1-2x).$

Consequently, the zero fixed point is stable for $0$x_n\in[0,1]$0. At $x_n\in[0,1]$1, stability is exchanged between the two fixed points. At $x_n\in[0,1]$2, the nonzero fixed point loses stability through a flip, or period-doubling, bifurcation.

The principal sequence of regimes is:

1. Absorbing or extinction regime: for $x_n\in[0,1]$3, trajectories converge to $x_n\in[0,1]$4.
2. Stable fixed-point regime: for $x_n\in[0,1]$5, typical trajectories converge to $x_n\in[0,1]$6.
3. Stable period-two regime: for $x_n\in[0,1]$7, the attracting state alternates between

$x_n\in[0,1]$8

1. Higher period-doubling regimes: stable period-four, period-eight, and higher-period orbits occur as $x_n\in[0,1]$9 increases.
2. Onset of chaos: the period-doubling cascade accumulates near

$n$0

1. Chaotic and periodic-window regimes: for larger $n$1, chaotic dynamics coexist with periodic windows, including a prominent period-three window near $n$2.

A fixed point is locally stable when $n$3. More generally, the multiplier of a period-$n$4 cycle is the derivative of the $n$5-fold iterate along the cycle. Stability is lost at multiplier $n$6 at an onset and at multiplier $n$7 at a period-doubling bifurcation.

The logistic map can also be written in the shifted symmetric form

$n$8

The transformation between the conventional and shifted parameters is

$n$9

A related quadratic representation is

$r$0

These representations describe equivalent quadratic-map dynamics but use different parameter conventions. Confusing $r$1, $r$2, and $r$3 can lead to incorrect comparisons of bifurcation values. The algebraic structure of periodic cycles and their stability boundaries has been derived through cyclic-polynomial reduction, characteristic determinants, and Möbius inversion (Zhang, 2012).

2. Period doubling, universality, and invariant measures

The successive period-doubling parameters accumulate according to the Feigenbaum scaling law,

$r$4

For successive bifurcation parameters $r$5, the ratios of parameter intervals approach $r$6. The constant is universal for broad classes of smooth one-dimensional unimodal maps with a single quadratic maximum. The universality concerns asymptotic scaling, not identical physical amplitudes, parameter values, or microscopic mechanisms.

The associated spatial scaling is governed by the Feigenbaum coordinate-scaling constant,

$r$7

At superstable period-$r$8 parameters, the lengths of orbit diameters are organized approximately according to binomial coefficients,

$r$9

producing approximate Pascal-triangle patterns. An analogous organization occurs for the widths of chaotic bands during the band-splitting cascade. These Pascal structures concern multiplicities of approximately equal geometric lengths; they are distinct from the parameter scaling governed by $[0,1]$0.

At the accumulation point, spatial positions and orbit subsequences scale approximately as inverse powers of $[0,1]$1. The Feigenbaum renormalization-group fixed-point map satisfies

$[0,1]$2

The coexistence of approximate binomial organization and power-law spatial scaling has been interpreted as producing both Gaussian-like and power-law statistical structures at the onset of chaos (Velarde et al., 2015).

The long-term behavior of the map is described by invariant probability measures. In regular regimes, the invariant measure may be atomic:

• a stable fixed point gives a single Dirac mass;
• a stable period-$[0,1]$3 orbit gives a finite combination of Dirac masses;
• chaotic regimes may produce singular, fractal, or absolutely continuous invariant measures.

At $[0,1]$4, the invariant density is the arcsine law,

$[0,1]$5

The density is absolutely continuous but has integrable inverse-square-root singularities at $[0,1]$6 and $[0,1]$7. Invariant densities are fixed points of the Frobenius–Perron operator. For a density $[0,1]$8, the operator satisfies

$[0,1]$9

The same invariant measures appear as asymptotic spectral densities in the Hermitian matrix logistic map. If

$0\leq r\leq4$0

then the eigenvectors remain fixed and each eigenvalue evolves independently according to the scalar logistic map. Consequently, a continuous initial eigenvalue density converges to the scalar invariant measure, which may be atomic, fractal, or absolutely continuous depending on $0\leq r\leq4$1 (Pawela et al., 2023).

3. Algebraic cycles and the Conley-graph structure

Every point of an $0\leq r\leq4$2-cycle is a fixed point of $0\leq r\leq4$3, although fixed points of $0\leq r\leq4$4 may have smaller periods dividing $0\leq r\leq4$5. Directly solving $0\leq r\leq4$6 becomes impractical because its degree grows as $0\leq r\leq4$7. The cyclic-polynomial method instead constructs a basis of square-free cyclic polynomials and reduces powers using

$0\leq r\leq4$8

If $0\leq r\leq4$9 is the matrix representing multiplication by the cycle multiplier in this basis, the characteristic equation is

$x=r x(1-x),$0

The determinant includes cycles whose periods divide $x=r x(1-x),$1. Möbius inversion removes shorter-period repetitions and produces primitive cycle polynomials. The number of cyclic basis elements is the number of binary necklaces,

$x=r x(1-x),$2

whereas the number of primitive period-$x=r x(1-x),$3 cycles is

$x=r x(1-x),$4

The method yields exact algebraic conditions for onset and bifurcation parameters through period $x=r x(1-x),$5. Examples in the conventional $x=r x(1-x),$6 parameter include:

• fixed-point onset: $x=r x(1-x),$7;
• first fixed-point bifurcation: $x=r x(1-x),$8;
• period-two bifurcation: $x=r x(1-x),$9;
• original period-three onset: $x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$0;
• original period-four onset: $x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$1;
• period-four descendant bifurcation: $x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$2.

The distinction between original cycles and period-doubling descendants is essential. An onset polynomial may contain factors associated with cycles born from shorter cycles and factors associated with genuinely original cycles. The algebraic framework separates these contributions and identifies the relevant real roots (Zhang, 2012).

A different structural description is supplied by the Conley graph. Chain-recurrent points are partitioned into nodes, where two points belong to the same node when each can be reached from the other by arbitrarily accurate pseudo-orbits. The directed graph of nodes contains an edge when a complete trajectory has one node in its backward limit set and another in its forward limit set.

For the logistic family, the Conley graph is always a tower: every pair of distinct nodes is connected by exactly one directed edge, and directed cycles cannot occur. The ordering is determined by distance from the critical point $x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$3,

$x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$4

Nodes farther from the critical point lie upstream from nodes closer to it. The upper nodes are repelling, while the bottom node is attracting. Nodes may consist of periodic orbits, repelling Cantor sets, attracting periodic orbits, trapping intervals, or attracting Cantor sets.

The tower is infinite precisely when the attractor is an almost-periodic, adding-machine Cantor set. This occurs on the parameter set associated with Feigenbaum-type Cantor attractors, including the Feigenbaum–Myrberg parameter near $x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$5. Periodic windows generate nested trapping regions and additional tower levels. The Conley description therefore includes recurrent structures that are invisible when only typical forward trajectories are examined (Leo et al., 2020).

4. Generalizations: deformation, nonlinear growth, memory, and stochasticity

Several generalized logistic maps modify either the nonlinear term, the growth law, or the temporal update mechanism.

A $x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$6-deformed map replaces the effective population variable by the $x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$7-number

$x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$8

and deforms both factors of the logistic nonlinearity:

$x_0^*=0,\qquad x_1^*=\frac{r-1}{r}.$9

The limit $f_r'(x)=r(1-2x).$0 recovers the standard map. For $f_r'(x)=r(1-2x).$1, especially $f_r'(x)=r(1-2x).$2, the graph becomes asymmetric and develops concave regions associated with rapid growth and slower decay. The trivial fixed point remains $f_r'(x)=r(1-2x).$3, while the nontrivial one becomes

$f_r'(x)=r(1-2x).$4

The first two-cycle bifurcation still begins at $f_r'(x)=r(1-2x).$5, although the locations of the branches depend strongly on $f_r'(x)=r(1-2x).$6. The Lyapunov exponent shows that changing $f_r'(x)=r(1-2x).$7 alone can move the system between regular and chaotic regimes (Banerjee et al., 2010).

Another generalization replaces the linear growth coefficient $f_r'(x)=r(1-2x).$8 by a hyperbolic-tangent function,

$f_r'(x)=r(1-2x).$9

The map is

$0

The parameters $0

$0

and its local stability is determined by

$0

Small changes in $0$x_n\in[0,1]$00 and chaos appears near $x_n\in[0,1]$01, substantially earlier than in the standard map (Delcourt, 2017).

Fractional logistic maps replace the instantaneous update by a history-dependent update. In Caputo formulations, the memory kernel is typically a power law,

$x_n\in[0,1]$02

whereas fractional-difference formulations use gamma-ratio or falling-factorial kernels. A representative Caputo fractional logistic map is

$x_n\in[0,1]$03

The fixed points remain $x_n\in[0,1]$04 and $x_n\in[0,1]$05, but stability depends on the memory kernel rather than solely on the ordinary derivative. Convergence is generally algebraic, often of order $x_n\in[0,1]$06, and apparent periodic behavior is more accurately described as asymptotically periodic. Long transients can produce cascade-of-bifurcations-type trajectories, inverse cascades, and strong dependence on the initial condition. Fractional versions exhibit numerical evidence of Feigenbaum scaling, but a complete theory of fractional chaos and universality remains open (Edelman, 17 Mar 2025).

The economic logistic map with memory is derived from a fractional accelerator equation and impulsive price shocks. Caputo derivatives produce Volterra equations whose discrete analogues retain the influence of all previous crises. When the fractional order is $x_n\in[0,1]$07, the memory kernel vanishes and appropriate rescaling recovers the ordinary logistic map. The model therefore interprets the standard map as a memoryless limit of an economic system with competition, power-law memory, and crisis impulses (Tarasova et al., 2017).

Randomness can be introduced by resampling the growth parameter,

$x_n\in[0,1]$08

where the $x_n\in[0,1]$09 are independent random variables. In the stable fixed-point regime, parameter noise decreases the long-term mean relative to the deterministic map at the mean parameter. In the stable period-two regime, sufficiently small parameter noise increases the mean. Numerical evidence suggests that the direction may alternate across successive period doublings, motivating a “flip-flop” conjecture for stable periods $x_n\in[0,1]$10 (Cruz et al., 2022).

5. Coupled, disordered, and experimentally realized logistic maps

Coupled logistic maps use the logistic output of neighboring maps as an additional control mechanism. A common form is

$x_n\in[0,1]$11

where $x_n\in[0,1]$12 is an adjacency matrix, $x_n\in[0,1]$13 is the degree of node $x_n\in[0,1]$14, and $x_n\in[0,1]$15 is the coupling strength.

Coupling can stabilize periodic behavior even when isolated maps are chaotic, produce synchronization, and generate multistability and hysteresis. In six-map electronic experiments with nominal parameters near $x_n\in[0,1]$16, increasing coupling generally produces a reverse sequence from chaos toward periodic and nearly synchronous states. Fixed points appear for approximately $x_n\in[0,1]$17. Hysteresis occurs because increasing and decreasing coupling sweeps can remain on different attracting sets. It persists across 52 network configurations despite parameter mismatch and electronic noise (Gutiérrez et al., 2020).

Electronic implementations use an analog multiplier to realize the quadratic nonlinearity and sample-and-hold circuits to create discrete-time evolution. In one implementation, the voltage correspondence is

$x_n\in[0,1]$18

with an AD633 multiplier and LF398 sample-and-hold components. Experimental bifurcation diagrams, Lyapunov exponents, periodicity charts, and phase portraits agree closely with numerical simulations. The architecture can be extended to coupled maps and larger networks (L'Her et al., 2015).

Disordered coupling changes the dynamical mechanism more radically. A system of $x_n\in[0,1]$19 maps with quenched random interactions is

$x_n\in[0,1]$20

where the random couplings have zero mean and variance $x_n\in[0,1]$21. In the large-$x_n\in[0,1]$22 limit, dynamic mean-field theory replaces the interaction term by a self-consistent Gaussian process. The fixed point becomes heterogeneous, with a truncated Gaussian distribution of component values and delta-function masses at the lower and upper cutoffs.

The Jacobian is a non-Hermitian random matrix. Its eigenvalue spectrum can leave the unit disk through:

• $x_n\in[0,1]$23, producing a May–Wigner instability governed by

$x_n\in[0,1]$24

• $x_n\in[0,1]$25, producing an oscillatory instability governed by

$x_n\in[0,1]$26

For $x_n\in[0,1]$27, arbitrarily weak disorder destabilizes the fixed point through the oscillatory channel. The resulting transition is generally a direct onset of high-dimensional chaos rather than a conventional sequence of two-, four-, and eight-cycles. Sufficiently strong homogeneous coupling can produce an isolated coherent eigenvalue resembling the scalar logistic multiplier and temporarily restore elements of the period-doubling cascade, but the disordered bulk eventually produces high-dimensional chaos (Baron et al., 21 Sep 2026).

The matrix logistic map extends the scalar system to Hermitian matrices,

$x_n\in[0,1]$28

Its eigenvalues evolve independently according to the scalar map, so matrix order controls the number of parallel scalar trajectories rather than changing the scalar rule. A broader form,

$x_n\in[0,1]$29

with $x_n\in[0,1]$30, generates coupled multidimensional logistic dynamics. When $x_n\in[0,1]$31 is associated with a graph, increasing coupling can transfer chaos from one vertex to others according to graph distance and connectivity. The same construction produces fractal spectral distributions, non-Hermitian singular-value dynamics, and complex eigenvalue clouds determined through free-probability relations (Pawela et al., 2023).

6. Computation, applications, and conceptual limitations

The logistic map is used in physical visualization, cryptography, reservoir computing, and models of gain–loss systems.

A microcontroller implementation divides $x_n\in[0,1]$32 into ten intervals and assigns one LED to each interval. Fixed points appear as a repeatedly illuminated LED, periodic cycles as repeating LED patterns, and chaotic trajectories as apparently irregular sequences visiting many bins. The coarse representation cannot resolve all higher-period cycles, because distinct numerical states may occupy the same LED interval (Serna et al., 2011).

In a hybrid cryptographic construction, the logistic parameter is used as a secret key together with a Hill cipher and an 8-bit nonlinear feedback shift register. The scheme reports successful file encryption and decryption, but its plaintext sensitivity is approximately $x_n\in[0,1]$33–$x_n\in[0,1]$34 and its key sensitivity approximately $x_n\in[0,1]$35–$x_n\in[0,1]$36, below the ideal $x_n\in[0,1]$37 avalanche level. The stated key-space calculation is internally inconsistent, and identifiability does not increase the number of candidate keys or prevent exhaustive enumeration. Chaotic sensitivity alone therefore does not establish modern cryptographic security (Mishra et al., 2012).

In reservoir computing, logistic iterations provide virtual nodes. An input is mapped to a parameter $x_n\in[0,1]$38, the map is iterated to generate a feature vector, delayed feature vectors provide temporal memory, and only a linear readout is trained. The method has been applied to a seventh-order polynomial and to Lorenz, Rössler, and Hindmarsh–Rose systems, including noisy data and self-prediction. Its effective dimensionality comes from parameter encoding, time multiplexing, nonlinear feature generation, and explicit delayed-state concatenation rather than from the state dimension of a single logistic map (Arun et al., 2024).

The logistic map also functions as a normal form for gain–loss systems. If gain increases with energy and nonlinear losses become stronger at high energy, their composition can produce a one-dimensional unimodal map. Under an appropriate homeomorphism, a physical energy map may be represented as

$x_n\in[0,1]$39

The equivalence is topological and dynamical rather than necessarily an exact algebraic identity. It preserves qualitative orbit structure and bifurcation organization, while physical amplitudes, time scales, and derivatives remain system-specific. Applications discussed include mode-locked lasers, rotating detonation engines, bouncing droplets, passive walking, sloshing, Faraday waves, and related damped-driven systems (Kutz et al., 2022).

Finite precision introduces additional dynamics. Iterating the map with a fixed decimal representation can create fixed and periodic patterns in least-significant digits even when the real-valued orbit is chaotic. For reported stable digit blocks, the periods beyond a fixed block follow

$x_n\in[0,1]$40

with

$x_n\in[0,1]$41

after the first non-fixed position. These effects are representation-dependent and should not be confused with periodicity of the real-valued logistic orbit (Valle et al., 2022).

Information-theoretic diagnostics characterize invariant-density structure through Fisher information and Cramér–Rao complexity. Fisher information is large for concentrated regular measures and generally decreases as the invariant distribution spreads, while the Cramér–Rao product

$x_n\in[0,1]$42

combines local roughness with global variance. A related effective temperature is defined from squared orbit increments,

$x_n\in[0,1]$43

These quantities are descriptive numerical diagnostics rather than thermodynamic invariants. Their values depend on finite sampling, histogram resolution, transient treatment, and endpoint conventions (Gomez et al., 3 Dec 2025).

The logistic map’s conceptual importance therefore extends beyond its original growth interpretation. It is simultaneously:

• a canonical one-dimensional route to chaos;
• an algebraically analyzable family of periodic cycles;
• a source of universal scaling laws;
• a normal form for gain–loss instabilities;
• a component of coupled and disordered high-dimensional systems;
• a testbed for stochastic, fractional, and memory-dependent dynamics;
• a nonlinear feature generator for reservoir computing;
• and a model in which finite precision, invariant measures, and information-theoretic structure can be examined explicitly.

Its familiar period-doubling cascade is robust within coherent one-dimensional unimodal families, but it is not generic under arbitrary modifications. Memory changes the update law and produces algebraic relaxation; noise changes invariant averages; coupling creates synchronization, hysteresis, and collective attractors; disorder can replace the scalar cascade with high-dimensional chaos; and finite representations can generate digit-level regularities absent from the underlying real dynamics.