Where Does a System Want to Go?

Attractors encode what a system "wants to be." Feedback loops are how it creates its own destination. But multistability reveals that possibility is never singular.

An attractor is a set of states toward which a system tends to evolve. Give the system a wide variety of starting conditions, and it will flow toward the attractor. This isn't metaphor — it's mathematics.

Formally, an attractor A is forward-invariant (once you're in A, you stay in A) and it attracts an open set of initial conditions (the basin of attraction). The system has somewhere it's going. That somewhere might be a single point, a repeating loop, or a fractal structure. But it has a destination encoded in its dynamics.

What Feedback Creates

Feedback is the mechanism by which attractors form.

A feedback loop is a circuit: outputs become inputs. The system folds back on itself. x(t) influences x(t+1) which influences x(t+2), but eventually the later states feed into the earlier ones. This circularity creates stability or instability depending on the sign.

Negative feedback is self-correcting. Deviation from a reference point triggers a restoring force. If x goes up, the system pulls it down. The classic example is a thermostat: temperature rises above setpoint, cooling activates, temperature falls. The equilibrium is a fixed point attractor. Homeostasis, cruise control, body temperature regulation — all negative feedback stabilizing around a target.

Positive feedback is self-reinforcing. Deviation triggers more deviation. If x goes up, the system pushes it higher. This can drive to extremes — runaway heating, microbial growth in rich medium, audio feedback howl. But positive feedback doesn't always explode. It can also create bounded loops and strange structures.

The Lorenz attractor emerges from three coupled differential equations with positive feedback. The result isn't chaos in the colloquial sense — it's deterministic chaos. The system loops forever through a butterfly-shaped region, never repeating exactly, but constrained to a specific fractal geometry. Run the system a thousand times from similar initial conditions and you get a thousand subtly different trajectories — all living in the same attractor.

Three Kinds of Destinations

Attractors come in types:

Fixed points. The system settles to a single state and stays there. A damped pendulum comes to rest. A stirred cup of coffee eventually stops moving. The simplest attractor, and the most common in engineered systems designed for stability.

Limit cycles. The system repeats a periodic orbit. Heartbeats, circadian rhythms, predator-prey population cycles. The system doesn't settle to a point — it settles to a loop. Stability lives in repetition, not rest.

Strange attractors. The system is confined to a fractal structure but never repeats. Lorenz, Rössler, Hénon — these systems are deterministic but unpredictable. Sensitive dependence on initial conditions means that arbitrarily close starting points diverge exponentially. The attractor constrains where the system can go, but within that constraint, anything can happen.

When Multiple Destinations Coexist

Here's where it gets interesting: a system can have multiple attractors.

Multistability means that different initial conditions flow to different stable states. Same system, same rules, different outcomes. The phase space is partitioned into basins of attraction — watersheds for where trajectories end up.

Bistability is the simplest case: two stable equilibria. A light switch is bistable — on or off, nothing in between. But multistability in dynamical systems is richer than mechanical switches. The basins can have complex geometry. Boundaries between them can be fractal. Small perturbations near a boundary can tip the system into a different attractor.

Cell differentiation is multistability in action. Every cell in your body has the same genome, but liver cells and neurons and muscle cells are stable in different gene expression patterns. The system has multiple attractors. Which one it settles into depends on initial conditions and signals received during development.

Hysteresis is multistability with memory. The system's current state depends on its history, not just current parameters. A magnetic material can be magnetized in either direction — which one depends on which field was applied last. The system remembers.

Where Bifurcation Enters

Bifurcation theory tells us how attractor structures change as parameters shift.

A bifurcation occurs when a small smooth change to a system's parameters causes a sudden qualitative change in behavior. The classic example: increasing a control parameter past a threshold, and a stable fixed point gives birth to a limit cycle. Or one fixed point splits into two. Or a stable point becomes unstable.

Poincaré named this "bifurcation" in 1885, identifying the phenomenon where stability itself shifts. The system was settled, then suddenly it isn't. The attractor structure undergoes phase transition.

Local bifurcations can be analyzed through changes in eigenvalues — when the real part of an eigenvalue crosses zero, stability is lost. Saddle-node bifurcations create or destroy fixed points. Pitchfork bifurcations split one equilibrium into three. Hopf bifurcations birth limit cycles from fixed points.

Global bifurcations involve larger invariant sets colliding. Homoclinic bifurcations: a limit cycle collides with a saddle point. Heteroclinic bifurcations: cycles connect multiple saddle points. The blue sky catastrophe: a limit cycle vanishes into non-existence as a parameter shifts.

What This Reveals

Attractors encode possibility. They show what a system can become, not what it must become.

Feedback loops create attractors. Negative feedback stabilizes; positive feedback amplifies. Together they sculpt the landscape of possibilities — valleys where the system settles, ridges where it's pushed away.

Bifurcations show that possibility itself has structure. Change a parameter slightly, and the set of available outcomes shifts fundamentally. The system that existed before the bifurcation is not the same system after, even though the difference is just a number.

Multistability reveals that the future isn't singular. Given the same rules, different starting conditions can flow to radically different endpoints. History matters. Initial conditions matter. Small perturbations at critical moments can tip a system into an entirely different basin.

The attractor is the memory of dynamics. The system doesn't know where it's going — it just follows local rules. But those local rules, iterated over time, encode a destination. The attractor exists as a mathematical structure before the system ever reaches it. It's what the system would become if it had infinite time.

This is the strange insight: the destination exists before the journey. The attractor describes all possible journeys' endpoints. Feedback loops are how the system approaches what it already, mathematically, is becoming.