A trajectory tells you where the system went. A vector field tells you what the system would tend to do from every state it might occupy.
Imagine a weather map covered with arrows.
At each location, one arrow tells you wind direction and speed.
The arrows do not show one aircraft’s journey.
They show the local motion available everywhere.
That is the intuition behind a vector field.
Quick Route
- State space: the field of possible states.
- Manifold: the structured subset the system occupies.
- Vector field: the local direction and rate of change assigned to each state.
- Trajectory: the path generated by following those local directions over time.
- Attractor: a state or set toward which the field directs nearby trajectories.
Canonical Job
Vector field owns one reader job in Cognitive Art:
Given the system’s current state, what direction and rate of change does the governing dynamic prescribe next?
In continuous dynamical systems, a vector field assigns a velocity vector to every point in state space.
Follow those local vectors and a trajectory emerges.
The 2023 Nature Reviews Neuroscience article Reconstructing Computational System Dynamics From Neural Data With Recurrent Neural Networks explicitly frames neural dynamical reconstruction in terms of state spaces, vector fields, fixed points and trajectories.
One-sentence answer: A vector field is a rule that assigns each state a local direction and rate of change, thereby specifying how trajectories flow through state space.
Vector Field Is Not Trajectory
The Cognitive Art article What Is a Trajectory? owns the path actually travelled.
Vector field owns the local rule that would generate paths from many starting states.
Trajectory:
this is what happened from this initial condition.
Vector field:
this is how the system would tend to move from every state in the model.
Vector Field Is Not Gradient
The existing Cognitive Art article What Is a Gradient? owns directional change in a scalar field.
A gradient is one special kind of vector field: it points in the direction of steepest increase of a scalar quantity.
But not every vector field is a gradient field.
Rotational flows can circulate without arising from one global scalar potential.
This difference matters because not every dynamical system is “rolling downhill” on one energy landscape.
Vector Field Is Not Force
In mechanics, force can generate acceleration.
A vector field is more general.
It can represent:
- velocity,
- acceleration,
- electric field,
- probability flow,
- neural latent dynamics.
The object is mathematical structure, not one physical mechanism.
The Arrow at One State
Suppose a learner’s state is defined by:
- concept mastery,
- speed.
At one moment, the learner is high mastery, low speed.
A training policy might predict:
next, speed should rise faster than conceptual mastery.
That prediction is one local vector.
Move to another state and the direction may change.
The Whole Field
A vector field becomes powerful because it assigns local motion everywhere, not only at the observed point.
This creates counterfactual structure:
- If the system started here, what path would follow?
- If it were perturbed there, where would it go?
- Which states flow toward the same attractor?
- Where are the boundaries between different futures?
A field therefore compresses infinitely many possible trajectories into one local dynamical rule.
Fixed Points Are Where the Vector Vanishes
At a fixed point, the vector field is zero.
If the system arrives exactly there, the deterministic dynamics do not move it.
But zero velocity alone does not tell us stability.
A fixed point can be:
- stable,
- unstable,
- a saddle.
To know which, inspect the surrounding field.
Saddles Organise Choice Without Being Attractors
A saddle can attract along one direction and repel along another.
Such states can organise trajectories even though they do not ultimately hold the system.
This is a useful reminder:
important dynamical states are not always endpoints.
Local Linearisation
Near a point, a nonlinear vector field can often be approximated by a linear one.
This local approximation can reveal:
- stable directions,
- unstable directions,
- rates of contraction,
- rates of expansion.
But local success does not guarantee global accuracy.
Move farther away and curvature or bifurcation can change the field.
Vector Field and Manifold
The new Cognitive Art article What Is a Manifold? gives the constrained geometry.
The vector field tells us how motion is organised on that geometry.
If vectors point mostly tangent to the manifold, natural trajectories remain on it.
If off-manifold vectors point inward, perturbations relax back.
This pairing is central to modern neural-dynamics analysis.
Vector Field and Attractor
An attractor is visible through the field.
Nearby arrows point inward toward the attracting set.
Change the field and the attractor can weaken, move or disappear.
Vector Field and Control Parameter
A control parameter can reshape the vector field without being one of the state coordinates.
Temperature can change the flow of a physical system.
Neural excitability can alter the stability of network states.
This is why parameter changes can produce bifurcations.
Reconstructing Vector Fields From Data
In real neuroscience, the vector field is not usually observed directly.
Researchers observe noisy trajectories.
Then they infer a dynamical model whose field could have produced them.
Durstewitz, Koppe and Thurm’s Nature Reviews Neuroscience perspective highlights recurrent neural networks as flexible tools for reconstructing nonlinear latent dynamics, including fixed points, vector fields and bifurcations, from neural time series.
This makes field reconstruction an inverse problem.
Many candidate fields may fit limited data.
Perturbation and out-of-sample prediction are needed to discriminate them.
Non-Claim: One Pretty Flow Plot Is Not the Dynamics
A vector-field visualisation can be persuasive.
But it is generated from a model.
If the latent coordinates change, the apparent field changes.
If data are sparse, unobserved regions are extrapolation.
If hidden variables matter, the field can appear non-Markovian or inconsistent.
Fields require validation, not admiration.
Vector Field in Mathematics
In differential equations, a system written as dx/dt = f(x) defines a vector field f over state space.
The field determines the instantaneous velocity of the state.
Solutions are trajectories that follow the field.
Fixed points satisfy f(x) = 0.
Bifurcations occur when parameter changes reorganise this structure qualitatively.
Vector Field in Education: Use as a Model, Not a Claim
A teacher can imagine a learning field.
At one learner state, retrieval practice produces the strongest movement.
At another, explanation is required before retrieval can help.
The same intervention has different vectors from different starting states.
This is a useful instructional model.
It is not evidence of one literal neural vector field for school learning.
The State-Dependent Teaching Rule
Instead of asking:
Which teaching method is best?
Ask:
From this learner state, which intervention produces the most useful next movement?
That question is vector-field thinking applied cautiously to pedagogy.
Vector Field in Organisations
Suppose an organisation is represented by cash, demand and capacity.
At high demand and low capacity, the natural direction may be:
rising backlog, rising error, falling customer satisfaction.
At the same demand but higher capacity, the vector changes.
Strategy is partly an attempt to reshape the field, not merely move the current point.
Failure 1: Trajectory Equals Field
One observed path is treated as the law governing every possible state.
Repair: collect trajectories from multiple initial conditions.
Failure 2: Gradient Equals Every Flow
All dynamics are imagined as descent on one scalar landscape.
Repair: allow rotational and non-conservative fields.
Failure 3: Hidden Variables
The same observed state appears to have different next directions because an unrepresented variable matters.
Repair: redesign the state representation.
Failure 4: Field Extrapolation
Arrows are trusted in regions with no data.
Repair: mark observed versus extrapolated regions and perturb where safe.
Failure 5: Static Parameters
One vector field is assumed to govern all regimes.
Repair: test whether parameters or context reshape the dynamics.
Repair Path
- Define the state variables.
- Observe trajectories from multiple starting states.
- Estimate the local direction and rate of change.
- Locate fixed points and saddles.
- Check whether the field is stable across time and context.
- Perturb discriminating directions.
- Compare rival field models.
- Revise the representation if one state seems to imply incompatible futures.
The Vector-Field Audit
- Vector field over which state space?
- What does each vector represent?
- What determines vector magnitude?
- Which regions contain fixed points?
- Which directions are stable or unstable?
- Is the field inferred or directly specified?
- Where is it extrapolating beyond data?
- Which parameters reshape the field?
- Do observed trajectories follow the predicted flow?
- What perturbation would falsify the model?
Research Notes and Further Reading
For a modern dynamical-systems treatment of neural state spaces, vector fields and trajectories, see Durstewitz, Koppe and Thurm, Reconstructing Computational System Dynamics From Neural Data With Recurrent Neural Networks (Nature Reviews Neuroscience, 2023).
For broader neural population dynamics, see Computation Through Neural Population Dynamics.
Cognitive Art uses vector field in its standard mathematical sense when discussing formal dynamics, and as a marked analogy when applying the concept to education or organisations.
World Return
A vector field earns trust when trajectories from new initial states follow its predicted flow.
If the arrows say the system should turn left and the real trajectory repeatedly turns right, the field is wrong.
Final Thought: The Future Is Local Before It Becomes a Path
A system never sees its whole trajectory at once.
It occupies one state.
Then the local dynamics push it somewhere next.
Again.
Again.
A vector field is the grammar of those next moves: the local rule from which the larger story of motion is written.