The IPS-based theory of psychology

Psychology names hundreds of patterns: attachment styles, burnout, the drama triangle, narcissistic injury, cognitive dissonance, epistemic humility. This essay shows that every one of them is a reading on the Fractal Persistence Equation (FPE) at the level of a conscious agent. There is no separate psychological law running beneath the thermodynamics — there is one accounting identity, and psychology names the ways individual agents manage it.

Formal derivation: information_persisting_systems.md. Vocabulary: glossary.md. Scale bridge to sociology and geopolitics: fractal_layers.md.


1 The equation at the agent layer

\[ \mathcal{R}^{(L)} = \Psi\!\bigl(\mathcal{R}^{(L+1)}\bigr) \cdot \left[\frac{P_{in}^{(L)}\,\eta(I)}{\omega^{(L)}\,\mathcal{E}_{\Sigma}^{(L)}\,\bigl(1 + \mathcal{D}_{KL}^{(L)} + \Gamma^{(L)}\bigr)}\right] \cdot \Phi\!\bigl(\mathcal{R}^{(L-1)}\bigr). \]

Why a person is an IPS in the first place

Before substituting terms, it is worth being explicit about why the equation applies to a person at all, because the rest of the essay is just this step repeated for different patterns. A conscious agent satisfies the four clauses of Definition 2.1 in the formal paper:

  1. Boundedness. The agent has a Markov blanket \(\partial\Sigma\) — the skin and sensory-motor interface, plus the behavioural boundary of what the agent will and will not do. Conditional independence (2.1) holds: given what the agent observes and does, the internal state is statistically separable from the rest of the world.
  2. Drive. The agent is a non-equilibrium steady state: it imports free energy (food, sleep-restored alertness, attentional arousal) and exports entropy as heat, CO₂, and discarded information. Stop the fluxes and the macrostate spreads within minutes to days — exactly the \(\tau_d\) bound of Theorem 5.1.
  3. Internal model. The agent carries a generative model \(q_\mu\) over external causes — what psychotherapy calls the internal self-model (ISM), the working set of beliefs about self, others, and what closeness/action costs. Behaviour is well-described as gradient descent on variational free energy (2.2); this is the formal content of “people act on what they expect.”
  4. Identity preservation. The agent persists as the same coarse-grained pattern (a recognisable person with a continuous biography) over \(\tau \gg \tau_{\mathrm{relax}}\), where \(\tau_{\mathrm{relax}}\) is the neural/metabolic update cycle of seconds to minutes.

Because the four clauses hold, the agent is governed by (4.4): there is no separate psychological law beneath the thermodynamics, only the substitutions that translate each term into the vocabulary of a life. The table below is that translation, not an analogy.

At the agent layer (\(L\) = conscious person), the terms substitute as follows:

FPE term Agent-layer reading Concrete example
\(P_{in}\) Metabolic and attentional drive (tonic + phasic) Blood glucose, sleep-restored alertness, phasic arousal
\(\eta(I)\) Coupling efficiency: action-to-outcome ratio Assertive vs. scattered or suppressed boundary work
\(\omega\mathcal{E}_\Sigma\) Structural complexity × noise floor Bureaucratic over-commitment; noisy, unsafe environment
\(\mathcal{D}_{KL}\) Delusion divergence: gap between internal model \(q_\mu\) and observable reality Denial, rationalisation, sustained self-deception
\(\Gamma\) Structural fatigue: backlog of unresolved friction Unspoken resentments, unpaid relational debts
\(\Phi\) Substrate integrity: body, sleep, attachment capacity HPA tone, immune function, somatic load
\(\Psi\) Shelter coefficient: protection from enclosing nodes Family safety, employment stability, cultural coherence

When \(\mathcal{R} < 1\), the agent is consuming its own structure: anxiety, somatic symptoms, depression, relational collapse. The mechanism is the proof of Theorem 5.1, read at this scale: the usable income \(P_{in}\eta\) falls short of the required dissipation \(\omega\mathcal{E}_\Sigma(1+\mathcal{D}_{KL}+\Gamma)\), so the internal free-energy reservoir \(\mathcal{B}\) depletes at rate \(\dot{\mathcal B} \le -\delta P_{out}^{req}\). The “borrowed time” is not metaphorical — it is the bounded quantity \(\mathcal{B}/(\delta P_{out}^{req})\) in (5.1). The agent either pays the debit (update the model, resolve the friction) or the macrostate leaves its equivalence class within that window: breakdown, rupture, or, at the limit, death.

The two operations

Almost every effective psychological intervention reduces to two moves:

  1. Lower \(\mathcal{D}_{KL}\) — align internal model \(q_\mu\) with observed reality. Say what is true (to self first, then other, then world-model).
  2. Lower \(\Gamma\) — settle or exit unresolved friction. Name observable + controllable consequence; repair or decouple.

The thermodynamic reason these two — and only these two — recur is that they target the only two denominator terms the agent can actively reduce. \(\omega\mathcal{E}_\Sigma\) is largely fixed by biology and environment (you cannot will away your noise floor); \(P_{in}\) is bounded by metabolism and circumstance. But \(\mathcal{D}_{KL}\) and \(\Gamma\) are informational debts that the agent services either by paying the Landauer cost of an honest update once or by paying the continuous suppression cost forever.

This is the content of Theorem 5.2, and it is worth slowing down on. Each nat of delusion divides the lifetime budget by a factor of \(e\) (equation 5.2). The mechanism is that a wrong \(q_\mu\) forces the agent to act on predictions that mismatch reality, generating surprise on every update cycle; by the Crooks/Jarzynski identity each nat of surprise costs at least one nat of excess dissipation per cycle (3.9). The agent is therefore paying Landauer three times — once to act, once to be wrong, once to suppress the mismatch — instead of once to update. The honest update is a one-time erasure cost (\(k_B T\ln 2\) per bit, equation 3.4); suppression is that same cost charged every cycle, compounded by the \(\Gamma\) it generates because the unresolved mismatch keeps producing micro-frictions. Over any horizon longer than a single cycle, paying once is cheaper than paying forever. That is the entire reason therapy, apology, and confession “work” in FPE terms: they convert a recurring thermodynamic tax into a one-time settlement.

\(\Gamma\) — structural fatigue — is the backlog of micro-events the blanket has absorbed but not repaired (unspoken resentments, unpaid relational debts, unmade repairs). It enters the denominator additively and, because each unresolved item keeps generating fresh surprise, the unresolved set \(C\) grows quadratically in cost: \(n\) outstanding conflicts do not cost \(n\) times one conflict, they cost roughly \(n^2\), since each pair can co-trigger. Operation 2 (settle or exit) does not just remove one item; it removes the cross-terms it was generating.

Apologies, therapy homework, DBT skills, NVC, and conflict-resolution protocols are all implementations of these two operations. The persistence ratio is indifferent to vocabulary.

Aggression and empathy are not opposites

The naive moral reading treats aggression as bad and empathy as good. The FPE forces a different reading, because the two map to different factors of the same product rather than to opposite ends of one axis. Aggression is the numerator factor \(P_{in}\eta\) — the work that actively maintains the Markov blanket by enforcing what is inside and what is outside. Empathy is a denominator-lowering operation — it keeps the coupling \(\mathcal{D}_{KL}\) low by accurately modelling the other, and it keeps \(\Gamma\) low by absorbing repairable friction before it sets. Because they are different factors, zeroing either one collapses the product. The boundary that is never enforced dissolves (Theorem 5.3: a blanket that cannot be maintained is no longer a blanket); the coupling that is never honestly modelled becomes exponentially expensive (Theorem 5.2).

Pole FPE role Failure mode
Aggression (\(P_{in}\eta\)) Maintains the Markov blanket — enforces the boundary Suppression → energy turns inward; misallocation → attacks others’ \(\Phi\)
Empathy (\(\downarrow\mathcal{D}_{KL}\), \(\downarrow\Gamma\)) Keeps coupled models cheap Without boundaries → compassion fatigue; without honesty → placation raises dyadic \(\mathcal{D}_{KL}\)

The Warrior enforces the boundary (\(S = P_{in}\eta\)). The Sage keeps the denominator low (\(T \propto 1/(\mathcal{D}_{KL}\,\Gamma)\)). Durable respect requires both factors non-zero. A doormat has \(S \approx 0\); a tyrant or a chronic people-pleaser has \(T \approx 0\) in practice.

The failure modes are asymmetric but symmetric in origin. Suppressed aggression does not vanish — the dissipation constraint (3.3) must still be paid, so the unspent \(P_{in}\eta\) is rerouted inward as sympathetic activation, rumination, and somatic load, attacking the agent’s own \(\Phi\) (this is the mechanism behind §4’s people-pleasing read). Misallocated aggression attacks others’ \(\Phi\), which by Theorem 5.3 degrades the substrate of the coupled system and, in repeat play, the agent’s own shelter \(\Psi\) as targets withdraw.


2 Substrate and attachment: what \(\Phi\) looks like

Substrate integrity is not abstract. It is sleep, HPA regulation, immune tone, and — because humans are co-regulatory mammals — attachment to reliable others.

The mechanism link is Theorem 5.3. The agent at level \(L\) is composed of sub-IPS at \(L-1\) (Section 2.5 of the formal paper): cardiovascular, immune, neural, microbiome, and — crucially for a social mammal — the attachment system that regulates arousal through coupled others. \(\Phi\) is the composition operator over these constituents: critical parts enter as a bottleneck (a \(\min\)), redundant parts as a softer pool. The theorem says the level-\(L\) system cannot persist if any structurally critical sub-IPS fails, regardless of how well the environment is buffered. “Get more sleep” is not folk wisdom bolted onto the theory; it is the substrate term of (4.4) — the cellular \(\mathcal{R}_i^{(L-1)}\) must itself be \(\ge 1\) or it drags the whole product down. This is also why no amount of numerator-side effort (“try harder”) can rescue a \(\Phi\) that is collapsing: \(\Phi\) is a multiplicative postfactor, and a product with a factor tending to zero tends to zero independent of the middle bracket.

Burnout and compassion fatigue

Burnout is routinely misframed as a numerator problem (“try harder,” “care more”). The FPE reads it as \(\Phi\) failure plus denominator overload across three layers. The misframing is not just unsympathetic — it is thermodynamically wrong, and acting on it accelerates the collapse. The mechanism: burnout is the joint failure of the postfactor \(\Phi\) and the middle bracket’s denominator, while the numerator is held fixed or squeezed. Squeezing a numerator against a collapsing \(\Phi\) is precisely the trajectory Theorem 5.1 flags as bounded-dissolution: \(\dot{\mathcal B} \le -\delta P_{out}^{req}\) with \(\delta\) growing as \(\Phi\) falls, because every unit of work now costs more of a shrinking reservoir.

Compassion fatigue is empathy without boundaries: empathy at high volume lowers dyadic \(\mathcal{D}_{KL}\) for others but continuously drains the denominator of the carer. When all three \(\Phi\) layers trend downward together while demands remain flat, burnout is the outcome. “Try harder” advice applied at this point is a numerator squeeze on collapsed substrate — the FPE-predicted path to accelerated dissolution.

Maslow’s hierarchy as nested \(\Phi\) and \(\Psi\)

Maslow’s pyramid (physiological → safety → belonging → esteem → self-actualisation) is a reading of the FPE dependency chain. Lower layers are \(\Phi\) preconditions for higher ones. The mechanism is fractal composition again: each Maslow level corresponds to a different \(L\) in (4.4), and the persistence ratio at the higher level multiplies the \(\Phi\) (and \(\Psi\)) of the level below. So the dependency is not a moral ladder but a multiplicative product — a single failing lower factor zeros the upper levels, exactly as Theorem 5.3 requires:

Maslow level FPE term Constraint
Physiological Cellular \(\Phi\) — glucose, sleep, temperature Without this, \(P_{in} = 0\); higher layers cannot run
Safety \(\Psi\) from shelter, income, predictable environment Reduced \(\mathcal{E}_\Sigma\); enables model-building beyond threat
Belonging Relational \(\Phi\) + \(\Psi\) from co-regulation Secure attachment lowers default \(\mathcal{D}_{KL}\) in coupling
Esteem Numerator efficiency: \(P_{in}\eta\) on goal-directed action Recognised agency; feedback that effort maps to outcome
Self-actualisation Running \(\mathcal{R} \gg 1\) — generativity, peak experience Surplus capacity routed to exploration and contribution

IPS sharpening. The hierarchy is not a ladder where each level is “finished before moving up.” It is a multi-scale \(\mathcal{R}\) stacking: you can temporarily run above-threshold on upper layers while cellular \(\Phi\) erodes (burnout trajectory). The FPE predicts that upper-layer “self-actualisation” purchased with depleted lower \(\Phi\) is not stable — the numerator looks large while substrate collapses. Maslow’s insight that most people never reach self-actualisation maps to: most environments impose enough \(\mathcal{E}_\Sigma\) and \(\Gamma\) to keep the denominator above the threshold where surplus exists.

Attachment styles as coupling defaults

Bowlby and Ainsworth described stable priors for how mammals regulate arousal in close coupling. FPE reads these as learned defaults for \(\Phi\), \(\mathcal{D}_{KL}\), and \(\Gamma\) in dyads. The mechanism is Bayesian: the attachment system is the part of the agent’s internal model \(q_\mu\) that predicts what closeness costs and what repair is available. Early caregiver responses set the prior on these quantities, and the prior becomes a default policy for how the agent manages the three denominator terms in future dyads. A secure prior predicts low-cost repair, so the agent can afford honest updates (low \(\mathcal{D}_{KL}\)) and prompt friction settlement (low \(\Gamma\)); an anxious prior predicts unreliable availability, so the agent over-monitors (high \(\mathcal{D}_{KL}\) that cannot be closed) and pursues at low \(\eta\); an avoidant prior predicts that closeness itself is costly, so the agent shrinks the coupled graph to keep \(\Gamma\) down at the price of co-regulatory \(\Phi\). Therapy that updates attachment is, formally, Bayesian model revision on this prior — paying the Landauer cost of rewriting \(q_\mu\) so that future closeness no longer triggers a prediction error it cannot afford:

Style FPE signature Collapse trajectory
Secure Predictable repair → small conflict set \(C\); accurate co-modelling → low \(\mathcal{D}_{KL}\); relational \(\Phi\) supports cellular \(\Phi\) Resilient
Anxious \(P_{in}\) mobilised but low \(\eta\) (protest, pursuit); continuous high \(\mathcal{D}_{KL}\) about partner availability Hyper-monitoring exhaustion; anxious–avoidant escalation spiral
Avoidant \(\Gamma\) reduced by shrinking the coupled graph; self-\(\mathcal{D}_{KL}\) (needs minimised) Stable until \(\Phi\) needs co-regulation — illness, grief, crisis
Disorganised Caregiver was both shelter and threat → incoherent \(\mathcal{D}_{KL}\) about closeness Highest allostatic load; simultaneous drive to raise and lower \(\Gamma\)

Secure attachment approximates fractal coherence: relational \(\Phi\) supports cellular \(\Phi\) across sleep, HPA tone, and immune function. Therapy that updates attachment is Bayesian model revision on the prior — lowering \(\mathcal{D}_{KL}\) about what closeness costs.


3 Delusion management: what \(\mathcal{D}_{KL}\) looks like

Delusion divergence is not a moral verdict; it is the thermodynamic gap between internal model and resolved reality. When that gap is non-zero, the agent either pays the Landauer cost to close it or suppresses surprise at continuous metabolic cost and rising allostatic load.

The mechanism worth stating plainly, because the rest of this section is just naming ways to avoid paying it: \(\mathcal{D}_{KL}\) enters the FPE twice — once in the numerator through the coupling efficiency \(\eta_I(\mathcal{D}_{KL}) \le 1 - \mathcal{D}_{KL}/\log|\mathcal{E}|\) (equation 4.2: a wrong model cannot extract work from gradients it cannot predict), and once in the denominator as the multiplicative tax \((1+\mathcal{D}_{KL})\) (equation 4.3, the delusion tax). Both hurt the ratio, and Theorem 5.2 quantifies the joint effect as an exponential lifetime penalty per nat. So delusion is not a soft, character-level flaw; it is a hard, double-counted thermodynamic loss. Every clinical pattern below is, mechanically, a different way of refusing to pay the one-time Landauer cost of an honest update and instead paying the recurring suppression tax — with the bill presented to \(\mathcal{B}\) and \(\Phi\) until a phase event (discovery, breakdown, rupture) forces settlement.

Denial, dissociation, cognitive dissonance

When reality \(P\) diverges from internal model \(Q\), three suppression strategies differ primarily in how they hide surprise. They are not three different psychologies; they are three bookkeeping choices for where to park the unmatched surprise so it does not (immediately) force a \(q_\mu\) update. Each keeps \(\mathcal{D}_{KL}\) nominally low on a local channel by paying the suppression cost elsewhere — on the agent’s own \(\mathcal{B}\), on a partitioned-off sub-model, or on a growing \(\Gamma\) residual. The FPE sees all three as the same refusal, just routed through different internal accounts:

All three share the same FPE signature: the agent is paying energy to avoid lowering \(\mathcal{D}_{KL}\) rather than spending that energy on the update itself. The debt accumulates until a phase event forces resolution — discovery, physical breakdown, relational rupture.

Why the debt must accumulate is the Crooks/Jarzynski identity (3.9): every cycle the mismatched channel produces surprise, and each nat of surprise carries an irreducible dissipation cost. Suppression does not delete this cost — thermodynamics forbids that — it merely routes it into \(\mathcal{B}\) (depleting the reservoir) and into \(\Gamma\) (the unresolved micro-frictions the suppression itself generates). The agent is, in effect, borrowing against \(\mathcal{B}\) at an interest rate set by Theorem 5.2. The phase event is simply the moment the loan is called: \(\mathcal{B}\) hits zero, the bound (5.1) saturates, and the macrostate leaves its equivalence class. This is why “the truth comes out” in collapses rather than gradually — the underlying quantity is a first-passage time on a depleting reservoir, not a smooth opinion change.

CBT cognitive distortions are named subtypes of \(\mathcal{D}_{KL}\) maintenance. Beck and Ellis catalogued them clinically; the FPE gives the thermodynamic account:

Distortion FPE mechanism Cost
Catastrophising \(\mathcal{D}_{KL}\) inflated over low-\(P\) outcomes; \(\mathcal{E}_\Sigma\) over-weighted Chronic sympathetic activation; \(P_{in}\) spent on threat-that-never-arrives
Mind-reading / fortune-telling Internal model \(Q\) assigned to others without evidence; high \(\mathcal{D}_{KL}\) that cannot be closed Relationship \(\Gamma\) rises from acted-on false attribution
All-or-nothing Binary \(Q\) over a continuous \(P\); no partial-credit update Large residual after any real outcome; \(\Gamma\) from perfectionism
Personalisation Causal attribution misdirected: external events written as self-caused \(\mathcal{D}_{KL}\) about agency; misallocated \(P_{in}\) on unfixable channel
Emotional reasoning Affective state taken as evidence about external \(P\) Secondary \(\mathcal{D}_{KL}\): emotion → confirmed false belief → more emotion
Should statements Normative \(Q\) imposed on stochastic \(P\); rigid \(\Gamma\) on self or other Continuous \(\Gamma\) from gap between demand and outcome

CBT works by exposing the discrepancy between \(Q\) and observable \(P\) through behavioural experiments and Socratic questioning — operation 1, lowering \(\mathcal{D}_{KL}\). Behavioural activation and exposure are operation 2 variants: close or bound the friction that the distorted model is avoiding.

Ego preservation vs. epistemic humility

Ego preservation is the stable attractor side of delusion management. When prediction error arrives, the agent faces two paths. The fork is the same Landauer-vs-suppression choice as §3, now stated as a policy rather than a one-off strategy: pay the erasure cost and reorganise \(q_\mu\) (humility), or refuse the erasure and pay the recurring suppression tax (ego preservation). The table below is the long-run accounting of each branch. The reason humility is the Sage function and not self-deprecation is that it is the cheaper bookkeeping in expectation — one erasure now versus compounding suppression forever, exactly as Theorem 5.2 dictates:

Path Short-term Long-term
Humility Pay Landauer cost: erase wrong bits, reorganise \(Q\) Lower \(\mathcal{D}_{KL}\); cheaper coupling; compounding trustworthiness
Ego preservation Keep \(Q\) stable; suppress surprise Rising \(\mathcal{D}_{KL}\); more suppression work; \(\Gamma\) explosion when truth leaks

Common ego-preservation moves and their FPE signatures:

Strategy FPE read
Rationalisation Patch \(Q\) without structural update
Deflection Export \(\Gamma\); keep local \(\mathcal{D}_{KL}\)
Identity fusion ISM and belief merged — update feels like death
Performative certainty Broadcast \(Q\) with higher confidence than internal precision supports
Moralisation Turn factual dispute into character attack; raise coupling \(\Gamma\)

Epistemic humility is not self-deprecation; it is calibrated bookkeeping: say what you know, what you infer, what you do not know. It is the Sage function — keeps trustworthiness high by keeping \(\mathcal{D}_{KL}\) and \(\Gamma\) low. Where feedback is delayed or punished, humility is not selected. That is a \(\Psi\) problem, not a character defect.

Emotions as FPE channel derivatives. Primary emotions are the qualia of deflection in named FPE channels; secondary emotions arise when two channels deflect together:

Primary FPE channel Secondary (co-deflection)
Surprise \(\mathcal{D}_{KL}\) spike Disappointment (+ sadness), Alarm (+ fear)
Fear \(\Psi\) drop Shame (+ sadness), Alarm (+ surprise)
Sadness \(\Phi\) / \(\Gamma\) Disappointment (+ surprise), Shame (+ fear)
Anger \(\eta\) blocked Contempt (+ disgust), Pride (+ joy)
Disgust \(\mathcal{D}_{KL}\) / hollow order Contempt (+ anger)
Joy \(P_{in}\eta\) surplus Pride (+ anger)

4 Friction dynamics: what \(\Gamma\) looks like

\(\Gamma\) is structural fatigue: the backlog of unresolved micro-events the Markov blanket has absorbed but not repaired. The mechanism that makes it grow quadratically rather than linearly is pairwise co-triggering. A single unresolved conflict \(c_i\) generates surprise on each update cycle at some rate \(\rho_i\). With \(n\) outstanding conflicts, the surprise is not \(\sum_i \rho_i\) but \(\sum_i \rho_i + \sum_{i\ne j}\rho_{ij}\), because conflict \(c_i\) and \(c_j\) can co-trigger (the same conversation re-opens both wounds; one resentment primes the next). The cross-terms dominate as the set \(C\) grows, which is why relationships often feel “suddenly” terminal — the cost curve is convex, and the phase transition in \(\mathcal{R}\) through 1 is sharp. This is also why early repair is wildly cheaper than late repair: removing item \(c_k\) from a set of \(n\) removes its \(O(n)\) cross-terms, not just its \(O(1)\) self-cost.

Escalation, Gottman’s horsemen, rupture–repair

Gottman’s Four Horsemen (criticism, contempt, defensiveness, stonewalling) are behavioural signatures of rising structural fatigue and delusion divergence in a coupled pair. Each horseman is, mechanically, a move that adds to \(C\) while appearing to discharge it — it converts an unresolved item into a new, larger unresolved item, which is why each cycle leaves the dyad further from \(\mathcal{R}\ge 1\) than before:

A delusion–friction loop: surprise and hurt → defensive ISM update → contempt and stonewalling → further model divergence. Each cycle adds to \(C\); structural fatigue grows roughly quadratic in the size of the unresolved set. Rupture–repair works by inserting the two operations: name the observable truth, then settle or exit the friction.

People-pleasing, codependency, fawning

People-pleasing looks like low aggression and high empathy. The FPE usually reads it as misallocated aggression: power income is not destroyed by boundary suppression — it is rerouted inward as anxiety, rumination, and somatic symptoms. When \(P_{in}\eta\) on the social boundary is chronically suppressed, the dissipation constraint is paid by the agent’s own substrate.

The mechanism is the conservation enforced by (3.3): the agent is a NESS, so \(\dot S_{int} \ge 0\) must be exactly exported. Boundary enforcement is one of the legitimate entropy-export channels — doing the hard, honest work of saying “no” or naming a consequence dissipates energy as structured action. When that channel is suppressed, the internal entropy production does not stop; it is rerouted. The unspent \(P_{in}\eta\) becomes sympathetic overactivation, rumination loops, and somatic symptom load — each a high-\(\mathcal{E}_\Sigma\), low-\(\eta\) form of dissipation that attacks \(\Phi\) directly. This is the same “aggression turns inward” mechanism as §1, now sustained chronically: the numerator looks preserved, but \(\eta\) on the boundary channel is near zero and the rerouted power taxes the postfactor. The FPE therefore predicts, counter-intuitively, that the chronic people-pleaser’s body fails before their social performance does — the substrate pays the bill the boundary refused to charge.

Codependency is the extended pattern: the agent’s model of self becomes entangled with the partner’s \(\mathcal{R}\) — “I am only okay if you are okay.” This exports the agent’s effective persistence condition to a system outside their control, dramatically raising denominator volatility. Formally, the agent has folded a level-\(L+1\) node (the partner) into their own identity equivalence class \([m_0]\), violating the level-distinction that keeps fractal composition stable (Theorem 5.3). Now the agent’s \(\mathcal{R}^{(L)}\) depends on another node’s \(\mathcal{R}^{(L+1)}\) that they cannot service — a substrate they do not control treated as if it were their own \(\Phi\).

Fawning is the threat-state variant: rational under high \(\Gamma\) risk (predatory partner, unsafe environment), it minimises the other’s \(\Gamma\) by absorbing it on own \(\Phi\). It is substrate sacrifice, not character defect. See §5 on dark-triad dynamics.


5 Stable attractors: common FPE pathologies

The term attractor is meant physically, not metaphorically. A psychological pattern is a stable attractor when the local FPE gradient pushes the agent back into the pattern faster than the two operations can push them out — so the configuration persists not because it is healthy but because, at the current \(\Psi\) and \(\Phi\), it is the local minimum of dissipation. The two patterns below are the canonical ones precisely because they are self-reinforcing: each move that avoids the two real operations (honest update, friction settlement) also lowers the short-run cost enough to lock the configuration in, while \(\Gamma\) grows underneath until a phase event breaks it. Understanding the mechanism is what distinguishes “this is a stable bad equilibrium” from “these people are just bad.”

The drama triangle

Karpman’s drama triangle (Persecutor, Victim, Rescuer) is a stable low-\(\mathcal{R}\) coupling attractor: each role avoids paying the full cost of boundary enforcement and honest model alignment, while exporting the cost into the structure. It is an attractor because the three roles form a closed loop of cost-export: each participant’s avoided operation becomes another participant’s \(\Gamma\), whose discharge becomes the first participant’s avoided operation again. No single member can exit the configuration by unilateral change, because the other two roles re-establish the equilibrium around them. This is the dynamical-systems content of “why drama is so hard to leave”:

Role What they avoid FPE exported
Persecutor Honest, low-\(\eta\) assertion on self \(\Gamma\) onto targets’ \(\Phi\)
Victim Phasic boundary enforcement \(P_{in}\) inward; \(\Gamma\) circulates
Rescuer Honest limits + friction resolution Hidden \(\mathcal{D}_{KL}\); resentment

Roles rotate — today’s Rescuer becomes tomorrow’s Victim when help is rejected — but the unresolved conflict set \(C\) is preserved through every rotation. The drama is cheaper than the two real operations in the short run; the \(\Gamma\) grows quadratically until the system collapses or one member decouples.

Why it is an attractor and not just a bad habit. Each role offers a local thermodynamic bribe. The Persecutor gets to feel effective without doing the low-\(\eta\) self-work of honest assertion (numerator looks high, real \(\eta\) is near zero). The Victim gets to discharge \(P_{in}\) inward as protest without the phasic boundary enforcement that would actually lower \(\Gamma\). The Rescuer gets the dopamine of apparent helpfulness while hiding the \(\mathcal{D}_{KL}\) they will not face in their own life. Each bribe is a short-run cut to the local cost surface, so the configuration is locally stable — exactly the sense in which “attractor” is meant. The cost is exported to the coupled \(\Phi\) and accumulates as \(\Gamma\) across the triangle, growing roughly as \(|C|^2\) as in §4. The collapse, when it comes, is therefore sharp rather than gradual, again because \(\mathcal{R}\) crosses 1 as a phase transition.

Interruption: name observables not identities (lower \(\mathcal{D}_{KL}\)); name consequence not identity attack (lower \(\Gamma\)); exit if the counter-party will not engage.

Dark triad and ASPD

Positive-sum vs. denominator-export coupling. In game-theory terms: healthy coupling is a positive-sum game; toxic coupling is negative-sum. In FPE terms, the same distinction is more precise. When all members run the two operations — honest \(\mathcal{D}_{KL}\) reduction and \(\Gamma\) settlement — coupling is positive-sum: each member’s \(\mathcal{R}\) rises as the others’ \(\mathcal{R}\) rises, because low \(\mathcal{D}_{KL}\) and \(\Gamma\) in the dyad pool compound through fractal composition (Theorem 5.3). The shelter \(\Psi\) grows stronger; relative persistence stays balanced because absolute persistence increases. The dark triad inverts this: the predator maintains local \(\mathcal{R} \geq 1\) by raising the denominator of coupled nodes — increasing its relative persistence within the system precisely because the system’s total persistence decreases. This is denominator export (negative-sum), not positive-sum coupling, and it cannot survive symmetric information.

The dark triad (narcissism, Machiavellianism, psychopathy) and clinical ASPD are not moral mysteries — they are fear-based local \(\mathcal{R}\) maintained by exporting denominator costs onto coupled nodes. The mechanism is the inverse of the two operations: instead of lowering one’s own \(\mathcal{D}_{KL}\) and \(\Gamma\), the dark-triad policy raises the denominator of coupled nodes so that the agent’s own ratio clears without doing the work. Concretely, gaslighting raises the target’s \(\mathcal{D}_{KL}\) (their model no longer tracks reality, so they cannot act effectively); contempt and rage raise the target’s \(\Gamma\) (unresolved friction directed at them); isolation lowers the target’s \(\Psi\) (no external shelter can corroborate reality or buffer shocks). Each move keeps the agent’s local \(\mathcal{R}\) above 1 by keeping the coupled node’s below 1. The policy is fear-based because it cannot survive symmetric information: under repeat play with enforcement and community memory (high \(\Psi\) for the target, transparent track records), the export stops working and the agent’s own denominator — long hidden — becomes due.

Trait Core FPE policy Collapse mode
Narcissism Grandiose \(Q\) (high self-\(\mathcal{D}_{KL}\) suppressed) + gaslighting others; rage defends ISM Supply loss, public exposure; supply holders’ \(\Phi\) exhausted
Machiavellianism Asymmetric \(\mathcal{D}_{KL}\) is the asset; \(\Gamma\) routed via third parties Audit, track-record transparency, alliance failure
Psychopathy Bold \(P_{in}\); low tonic cost of \(\Gamma\); export harm, decouple before costs return Repeat-play with enforcement; community memory
ASPD Chronic denominator export since youth Incarceration, age, isolation, exhausted \(\Phi\) in targets

Targets stay not because of stupidity but because of \(\mathcal{R}\) management under asymmetric information: intermittent reinforcement keeps target \(\mathcal{D}_{KL}\) uncertain about whether safety is coming; isolation cuts external \(\Psi\); entanglement raises exit cost. The mechanism is the target’s own FPE, held below 1 by design. Intermittent reinforcement is a precision-manipulation attack on the target’s \(q_\mu\): by keeping the predictive signal noisy, it forces the target’s model to maintain a high-variance posterior that cannot settle, which by (3.9) incurs continuous excess dissipation and by Theorem 5.2 shortens the target’s lifetime budget exponentially. Isolation removes the redundant \(\Psi_j\) channels (Section 2.4 of the formal paper) that would otherwise corroborate reality — once \(\Psi_{eff}\) collapses to the single channel the predator controls, the target has no external blanket against which to check their model. Exit cost is raised by folding the predator into the target’s \([m_0]\) (codependency, §4): leaving now feels like self-dissolution, because the level distinction has been collapsed. None of this requires the target to be irrational; each step is a locally rational response to a manipulated cost surface.

Fractal pattern: the same fear-based \(\mathcal{R}\) appears at every scale — controlling partner, narcissistic parent, charismatic executive, authoritarian state. What changes is which \(\Psi\) enables export and how hard exit is. The accounting is identical.

Interruption for the coupled target: lower your \(\mathcal{D}_{KL}\) by trusting observables over narrative; lower \(\Gamma\) primarily by decoupling (repair requires both parties — dark-triad attractors often treat repair as narcissistic injury). Restore \(\Phi\) first: sleep, somatic work, safe attachment elsewhere.


6 Voice, exit, and loyalty

Hirschman’s triad maps directly onto \(\Gamma\) and \(\mathcal{D}_{KL}\) management under changing \(\Psi\). The mapping is forced rather than imposed: voice, exit, and loyalty are simply the three policies available to an agent whose enclosing \(\Psi\) is degrading and who must decide whether to attempt the two operations inside the coupling or outside it. Voice is operation 1 + operation 2 attempted in place; exit is operation 2 (decouple) chosen because operation 1 is blocked or refused; loyalty is the deferral of exit, rational only when the deferred cost is recoverable. The table is the policy menu, not an analogy:

Strategy FPE When rational
Voice Attempt to lower \(\mathcal{D}_{KL}\) and \(\Gamma\) inside the coupling \(\mathcal{D}_{KL}\) is shared error; partner engages; your \(\Phi\) can sustain conflict cost
Exit Decouple to remove \(\Gamma\) Unresolved set \(C\) large, partner refuses repair; \(\mathcal{R}^{(L+1)}\) below viability
Loyalty Defer exit; often hides \(\mathcal{D}_{KL}\) \(\mathcal{R}^{(L+1)}\) is high or recoverable; short-term \(\Gamma\) spike buys long-term \(\Psi\)

Pathological loyalty is loyalty when \(\mathcal{D}_{KL}\) about the system is high and rising — staying to avoid exit cost while \(\Phi\) bleeds. The FPE is indifferent to attachment shame (“quitters fail”): leaving a collapsing \(\Psi\) is substrate preservation, not moral failure.

Voice fails when Persecutor dynamics punish honesty. Exit is the \(\Gamma\)-optimal move when: the unresolved set is large and partners refuse repair; \(\mathcal{R}^{(L+1)}\) falls below viability; voice raises your \(\mathcal{D}_{KL}\) without changing \(P\). Agents invested in their own persistence are structurally invested in the \(\Psi\) (relationships, institutions, biosphere) — undermining \(\mathcal{R}^{(L+1)}\) for private numerator gain is medium-term self-harm.


7 Diagnostic questions (any pattern)

For any psychological or relational pattern, ask:

  1. Numerator: Is \(P_{in}\) available (tonic + phasic)? Is it spent on the boundary or misallocated — inward, outward on others’ \(\Phi\), or wasted as low-\(\eta\) heat?
  2. Denominator: Where is \(\mathcal{D}_{KL}\)? Denial, deception, self-deception? How large is \(\Gamma\) (unresolved set \(C\) → roughly quadratic cost)?
  3. Fractal: Is \(\Phi\) intact (sleep, body, key attachments)? Is \(\Psi\) intact (job, family, culture, ecology)?
  4. Operations: Which of the two operations is blocked — honesty or friction-resolution? What would it cost to apply them?

8 What psychology is not in IPS terms


See also