Contact
Night sky over Joshua trees. Photograph by Renden Yoder.
03v1.0 · October 2026 · Arcas DynamicsSleepMemory
HypothesisNeuroscience

The lucidity trade-off.

REM sleep is credited with loosening memory and aiding insight; lucid dreaming, with offering a place to practise. We propose that the two pull in opposite directions. If prefrontal re-engagement during lucid REM narrows associative sampling toward strong traces, lucidity should help the rehearsal of what is already known and hurt the recombination that insight depends on.

Control helps a link only if that link is stronger than what is currently being sampled.

The argument.

From what REM sleep does to memory, to a prediction that can be tested within one person and one night.

  1. 01REM is a low-control statePrefrontal cortex is relatively quiet in REM sleep, and after REM awakenings weak associations are easier to reach than strong ones.
  2. 02Lucidity may restore controlLucid REM has shown more frontal gamma activity than non-lucid REM, although that evidence rests on very small samples.
  3. 03Control narrows samplingIn the model, prefrontal control acts as an inverse temperature: as it rises, retrieval concentrates on the strongest associations.
  4. 04A crossoverLucidity should help the rehearsal of strong traces and hurt recombination that depends on weak links, within the same person and night.

Live model.

The paper’s sampling model, live. Raise prefrontal control and watch where thirty draws from one cue land.

Fig. 3a Thirty draws from one cueLive model
β(c)3.00
Mean strength sampled0.440
Rehearsal, R(c)4.47
Recombination, X(c)0.335

At control c = 0.50, the rehearsed trace is reactivated 4.47 times and the chance of sampling the solution is 0.335.

Fig. 3b The crossoverConditional on incorporation
Rehearsal R(c), 0 to 15 reactivationsRecombination X(c), 0 to 0.5

Model estimates under illustrative parameters: K = 50 associates (one rehearsed trace at s = 1.0, four at 0.7, 45 at 0.2), β₀ = 0, β₁ = 6, M = 30.

The equations.

01 Control sets the temperature
β(c)=β0+(β1−β0) c\beta(c) = \beta_{0} + (\beta_{1} - \beta_{0})\,\htmlClass{hi}{c}

c is near 0 in non-lucid REM and near 1 in lucid REM with active dream control.

02 Sampling an associate
pk(c)=exp⁡(β(c) sk)∑jexp⁡(β(c) sj)p_{k}(c) = \frac{\exp\bigl(\beta(c)\, s_{k}\bigr)}{\sum_{j} \exp\bigl(\beta(c)\, s_{j}\bigr)}

A low β samples almost uniformly; a high β concentrates on the strongest links.

03 Rehearsal and recombination
R(c)=M prehearsed(c),X(c)=1−(1−q(c) psolution(c))MR(c) = M\, p_{\mathrm{rehearsed}}(c), \qquad X(c) = 1 - \bigl(1 - q(c)\, p_{\mathrm{solution}}(c)\bigr)^{M}

Expected reactivations of the practised trace, and the chance that a weak solution link is sampled at least once.

04 The boundary condition
∂pk∂β=pk (sk−Ep[s])\frac{\partial p_{k}}{\partial \beta} = p_{k}\,\bigl(s_{k} - E_{p}[s]\bigr)

Raising control increases access to link k if, and only if, its strength exceeds the mean strength being sampled.

What it builds on.

What sleep research already shows, as the paper cites it. The evidence is thin, and the paper says so.

g ≈ −0.06
effect of memory cues delivered in REM sleep across 15 experiments, after two outliers were excluded: no reliable benefitHu et al., 2020, Psychological Bulletin
About 40%
improvement on primed remote-associate problems in nap participants who reached REMCai et al., 2009, PNAS
1 of 9
puzzles solved after lucid dreams of them, against 6 of 13 after non-lucid dreams; small, post hoc and unit-levelKonkoly et al., 2026, Neuroscience of Consciousness
14 of 28
cued participants who reached signal-verified lucidity in a laboratory nap protocolCarr et al., 2023, Psychology of Consciousness

Predictions.

Each one names the result that would refute it.

  1. P1Rehearsal improvesRefuted if: items rehearsed in verified lucid REM are retained no better than items cued in non-lucid REM.
  2. P2Recombination suffersRefuted if: once a puzzle enters a dream, the solve rate after lucid episodes is equal or higher.
  3. P3Opposite directionsRefuted if: lucidity moves rehearsal and recombination in the same direction.
  4. P4Control, not awarenessRefuted if: an instruction to direct or release the dream has no effect, or awareness alone carries the cost.
  5. P5An associative signatureRefuted if: the strong-to-weak priming ratio after lucid REM matches that after non-lucid REM.
  6. P6A prefrontal indexRefuted if: lucid and non-lucid REM show no difference in frontal engagement.

How to test it.

First testWithin-person crossover

Three overnight sessions, with lucidity counted only when signalled by left-right eye movements during scored REM. Object locations are rehearsed and insight puzzles incubated in both states, with retests the next morning and a week later. About 51 recruits (model estimate).

P1P2P3P6
Experiment 2Direct or release

After a verified lucid signal, a randomised instruction either to direct the dream or to let it unfold, followed by strong and weak semantic primes and an anagram set on waking.

P4P5
Experiment 3Frontal stimulation

An exploratory arm: frontal 40 Hz stimulation or sham during cued REM, to test whether raising frontal engagement shifts outcomes toward rehearsal and away from recombination.

P6

Where it could be wrong.

Lucidity may not re-engage prefrontal cortex.

The premise rests on six participants and one case study. If Experiment 1 finds no frontal difference, the mechanism fails even if the crossover appears.

REM’s role in memory is uncertain.

Suppressing REM with medication did not impair skill memory, so lucid rehearsal may be no different from rehearsal while awake.

Recombination gains are small.

Cueing in REM did not improve problem solving in two studies, which leaves little room to detect a cost of lucidity.

Lucid dreams are hard to produce.

Induction is unreliable, and only a minority of participants may contribute lucid episodes at all.

About this paper.

Authors
Arcas Dynamics, Toronto, Canada
Version
1.0, October 2026
Predictions
6, none tested yet
DOI
Not yet assigned
Preprint
Not yet posted
Licence
CC BY 4.0. Share and adapt with credit.
Download PDF

Changelog

  1. v1.0 · October 2026First published.

How to cite

@misc{arcas2026lucidity,
  author       = {Arcas Dynamics},
  title        = {The lucidity trade-off},
  year         = {2026},
  month        = oct,
  note         = {Hypothesis paper 03, version 1.0, CC BY 4.0},
  howpublished = {\url{https://arcasdynamics.com/research/lucidity-trade-off}}
}