Part II: Geometry of Feeling

Dynamics and Transitions

0:00 / 0:00

Affect Trajectories

Affects are trajectories through affect space, not points:

dadt=F(a,o,a,context)+η\frac{d\mathbf{a}}{dt} = F(\mathbf{a}, \obs, \action, \text{context}) + \bm{\eta}

where a=(Val,Ar,Φ,reff,CF,SM)\mathbf{a} = (\valence, \arousal, \intinfo, \effrank, \mathcal{CF}, \mathcal{SM}) in the working basis. The force field FF has eigenskeletal structure: its Jacobian ∂F/∂a\partial F / \partial \mathbf{a} at each point has stiff and soft directions, and how those eigenspaces connect across the space is the eigenskeleton of the dynamics. Stiff directions are the dominant affect modes, the transitions the system is most likely to follow. Because the space is continuous, adjacent affects blend along them:

  • Fear →\to anger as causal attribution externalizes
  • Desire →\to joy as goal distance →0\to 0
  • Suffering →\to curiosity as valence flips while CF\mathcal{CF} stays high
  • Grief →\to nostalgia as arousal falls and CFapproach\mathcal{CF}_{\text{approach}} replaces CFavoidance\mathcal{CF}_{\text{avoidance}}

Attractor Dynamics

Some regions of affect space are attractors: once entered, the system tends to stay. A region R⊂A\mathcal{R} \subset \mathcal{A} is an attractor when the system is more likely to remain in it than to enter it from outside,

P(at+τ∈R∣at∈R)>P(at+τ∈R∣at∉R)\prob(\mathbf{a}_{t+\tau} \in \mathcal{R} \mid \mathbf{a}_t \in \mathcal{R}) > \prob(\mathbf{a}_{t+\tau} \in \mathcal{R} \mid \mathbf{a}_t \notin \mathcal{R})

for some characteristic time τ\tau.

Basin Geometry of Affect SpaceStable states are basins; transitions require crossing ridgesDepressionlow r_eff, high SM, neg VStable Neutralmoderate all dimensionsFlourishinghigh Φ, high r_eff, pos Vcrisis thresholdgrowth edgecurrent stateenergydeeper basin = more stable attractor
Affect Landscapebasin depth determines stability; transitions are causaldepressiondeep, narrowanxietyshallowcontentmentdeep, wideangercausal attribution externalizesactivation energy requiredstability →Affects are trajectories on this landscape, not static points

Two properties come apart in practice: position, where in affect space the system sits, and basin geometry, how stable the attractor is, its depth, width, and recovery rate. A system can hold a viable position inside a shallow basin, one perturbation from tipping into pathology, while another sits at a worse position inside a deep one. Contentment tracks basin geometry, not position: the felt sense that perturbations do not cascade, that the dynamics return, that the invariants you care about hold. Contentment is the phenomenology of a deep basin; anxiety is the phenomenology of a shallow one, viable but felt as precarious. A world of bliss is not maximal positive stimulation. It is a world where the invariants that matter, relational, material, self-model, are held by the environment with enough redundancy that defending them costs nothing.

Pathological attractors. Depression, addiction, and chronic anxiety are pathologically stable attractors:

  • Depression comes in two structurally different forms. Melancholic depression is a deep aversive attractor at low κ\kappa: the perceiver's modes have decoupled, so perception no longer feeds affect and the world is seen but not felt; low gain γ\gamma, so nothing new gets through the prior, which is the anhedonic quality; collapsed reff\effrank, the representation narrowed to one dwelt-upon dimension; low Ar\arousal; CF\mathcal{CF} collapsed, so felt alternatives are foreclosed; and high σattention\sigma_{\text{attention}}. The dynamics reliably return there. It is not the absence of a stable fixed point but the presence of a terrible one, and it is distinct from suffering, which is high-Φ\intinfo and vivid. Agitated depression is the opposite failure: no stable attractor at all. The system crosses shallow basins none of which holds, restless groundlessness rather than dead flatness. Both present clinically as depression and they need different work. The melancholic form needs coupling restored and the landscape restructured until viable attractors compete on stability. The agitated form needs a basin built first, any stable configuration, which can then be deepened toward viability.
  • Addiction: an attractor at high Val\valence conditional on the substance, with reff\effrank collapsing in goal space.
  • Anxiety: a diffuse attractor at low Val\valence, high Ar\arousal, high CF\mathcal{CF} spread across many threats.
  • Dissociation: the collapse of Φ\intinfo. The unified field fractures into subsystems processing on their own. Naive Lenia patterns do this under stress (ΔΦ=−6.2%\Delta\intinfo = -6.2\% in ). Falling apart is the cheap path. Integration that holds or rises under stress is the expensive achievement, and Part I's experiments show how rarely it is reached.

Identity consolidation and catastrophic forgetting. The attractor landscape is not fixed. It consolidates over development. Early in life basins are shallow and plastic, which is necessary for learning and is also a vulnerability: adversity or relational inconsistency early can consolidate pathological attractors before viable ones deepen. As development proceeds the landscape hardens around whatever has been traversed. Healthy consolidation produces several viable basins with navigable transitions, deep enough to contain normal variation and recover from moderate perturbation. Flexibility on the perceptual axes, the ability to move α\alpha, κ\kappa, and γ\gamma rather than sit locked, is in these terms a measure of between-basin navigability. Pathological consolidation takes two forms: one dominant basin with no exit, the melancholic pattern, identity calcified around one configuration; or a landscape that never reaches depth anywhere, the agitated pattern, consolidation never completed. 's stress-overfitting result is a substrate analog: patterns evolved under one stress regime become more integrated and more fragile, and collapse under novel stress that naive patterns handle better. The human version is an identity tuned to one developmental environment, a family dynamic, a class position, a cultural script, that works there and collapses under regime change. So durable therapeutic change is not repositioning inside a fixed landscape. It is restructuring the landscape: deepening viable basins, raising barriers around pathological ones, widening the transitions between healthy configurations. Insight alone does not do that. Repeated traversal under consolidating conditions does.

Force and Inertia in Identity Space

The opportunity deficit D=V−TD = V - T (above) defines a scalar field over identity space. The force on an identity is its gradient, the pull toward regions where the deficit narrows:

F(i,t)=−∇CD(i,t)F(i,t) = -\nabla_C D(i,t)

It has two components: an attractive force toward regions where traversal speed can rise, the pull of achievable goals, and a repulsive force away from regions where the landscape opens faster than any traversal could match, the vertigo of too much possibility. The mass of an identity is its resistance to a change in traversal direction, the inertia of accumulated commitments, relationships, and self-model structure. A high-mass identity is hard to accelerate and hard to deflect. A low-mass identity is plastic and uncommitted.

Some old words acquire structural meanings. Calling is a region where FF is strongly attractive. Purpose is a trajectory with consistently positive T/VT/V and aligned force. Despair, in Kierkegaard's sense, is high VV with near-zero TT and flat force: the landscape visible, vast, and no gradient to follow. Flow is T≈VT \approx V locally, traversing at the rate the landscape opens. Enlightenment, at least in the Buddhist formulation, reduces VV rather than raising TT: the landscape shrinks to what is present and OSR→1\text{OSR} \to 1 by releasing attachment to the untraversable.