Bistability and hysteresis in the proximity-based grouping of dot lattices Authors: Hamze Moazzen, Shahriar Gharibzadeh, Fatemeh Bakouie Published: 2025 ABSTRACT Grouping by proximity is a fundamental principle of human vision where elements close to each other are perceived as a group. This study examined the hysteresis effect in proximity-based grouping of dots in rectangular lattices in the presence of orientation biases. To induce perceptual switching, interdot distances in a specific direction were gradually increased or decreased on a trial-by-trial basis during ascending and descending sequences. To address orientation biases, stimuli were presented along two axes, each with two orientations. This design allowed investigation of perceptual switching between vertical and horizontal grouping in the cardinal axis and between 45° and 135° grouping in the oblique axis. Data from 34 participants were analyzed by fitting psychometric functions and the Lotka-Volterra-Haken (LVH) model—a dynamical model of neural population competition—with statistical analysis using linear mixed-effects models. The results revealed a strong preference for the vertical orientation in the cardinal axis and a slight bias toward 135° in the oblique axis. A significant hysteresis effect was found in both axes, with ascending transition points consistently exceeded descending points. Interestingly, this effect remained unaffected by stimulus axis or orientation, indicating robustness against orientation biases. A significant positive correlation in hysteresis across axes suggests it may represent a personal perceptual characteristic. The LVH model’s numerical simulation effectively captured the dynamic behavior of competing responses and their bifurcation during perceptual switching. The model demonstrated acceptable accuracy in estimating transition points and simulating individual responses. Moreover, its parameters reflected the observed data patterns and provided a mechanistic account of perceptual switching in proximity-based grouping.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-25575-3. Introduction Multistability in visual perception is a fascinating phenomenon where a single, unchanging stimulus can be perceived in multiple, alternating ways1,2. This occurs when the brain is presented with ambiguous figures that can be interpreted in two or more distinct percepts3,4. Classic examples include the Necker cube5 (Fig. 1A), which can be seen from different depth perspectives, and the Rubin vase6 (Fig. 1B), which can be perceived as either a vase or two faces. In both cases, the two percepts will alternate spontaneously; sometimes one is seen, sometimes the other, but never both at once1.Fig. 1Examples of multistable visual stimuli. (A) The Necker cube can be perceived from two different perspectives: either looking down on the cube from above (left-oriented) or looking up at it from below (right-oriented). (B) The Rubin vase can be perceived as either a vase or two faces in profile. (C) A group of equilateral triangles with the same rotation angle can be perceived as pointing to the right, top-left, or bottom-left. (D) A rectangular dot lattice is presented, with the main grouping organizations shown as vectors a–d. The perception of the lattice as a group of dots aligned in rows, columns, or diagonals exhibits multistability. Based on the proximity principle, smaller interdot distances increase the probability of grouping dots along that direction. Thus, grouping dots along vectors a and b is more likely than along vectors c and d, with the aspect ratio (AR =|b|/|a|) determining which grouping, a or b, is more probable to be perceived. Here, |a| =|b| yields an equal probability of dots being grouped in rows and columns. However, the prevalent preference for vertical orientation leads to a slightly higher probability of vertical grouping. (E) A lattice with significantly higher horizontal interdot distances (|b|= 1.5*|a|) is more probable to be perceived as a group of dots in columns. (F) A lattice with remarkably higher vertical interdot distances (|a|= 1.5*|b|) is more probable to be perceived as a group of dots in rows. Examples of multistable visual stimuli. (A) The Necker cube can be perceived from two different perspectives: either looking down on the cube from above (left-oriented) or looking up at it from below (right-oriented). (B) The Rubin vase can be perceived as either a vase or two faces in profile. (C) A group of equilateral triangles with the same rotation angle can be perceived as pointing to the right, top-left, or bottom-left. (D) A rectangular dot lattice is presented, with the main grouping organizations shown as vectors a–d. The perception of the lattice as a group of dots aligned in rows, columns, or diagonals exhibits multistability. Based on the proximity principle, smaller interdot distances increase the probability of grouping dots along that direction. Thus, grouping dots along vectors a and b is more likely than along vectors c and d, with the aspect ratio (AR =|b|/|a|) determining which grouping, a or b, is more probable to be perceived. Here, |a| =|b| yields an equal probability of dots being grouped in rows and columns. However, the prevalent preference for vertical orientation leads to a slightly higher probability of vertical grouping. (E) A lattice with significantly higher horizontal interdot distances (|b|= 1.5*|a|) is more probable to be perceived as a group of dots in columns. (F) A lattice with remarkably higher vertical interdot distances (|a|= 1.5*|b|) is more probable to be perceived as a group of dots in rows. Beyond these classic examples, a group of elements can also induce multistable perception. For instance, in the example shown in Fig. 1C, equilateral triangles with the same rotation angle can be perceived in three ways: as a group of elements pointing to the right, top-left, or bottom-left direction1. The multistability phenomenon reveals the dissociation between the objective properties of the physical stimulus and the subjective sensation2. Furthermore, it highlights the active role our brain plays in constructing our perception of the world4,7 and illuminates the relationship between conscious perception and brain functions8. One of the best-known visual stimuli subject to multistability is a dot lattice, a regular pattern of circular filled dots arranged in a two-dimensional plane9,10. In rectangular lattices, as shown in Fig. 1D, dots are aligned in the direction of two perpendicular vectors: a (in columns) and b (in rows). These lattices are multistable because they can be perceived as strips of dots grouped in rows, columns, or diagonals, but not simultaneously1,11. However, perception along the a and b vectors is dominant, while perception along the c and d vectors rarely occurs spontaneously12–14. Therefore, the perception of rectangular lattices is often considered bistable. In extreme conditions, rectangular lattices will be perceived as a group of dots in columns if the horizontal interdot distance is significantly greater than the vertical distance (Fig. 1E). On the other hand, these lattices will be perceived as a group of dots in rows if the horizontal interdot distance is considerably less than the vertical distance1 (Fig. 1F). These observations occur due to the principle of proximity, which states that objects closer together are more likely to be perceived as a group15. Pioneered by Wertheimer16, dot lattices have been used to quantify the principles of perceptual grouping, previously known as Gestalt laws17. By manipulating interdot distances in experimental trials and asking participants to report their perceived grouping, Kubovy, Wagemans, and their colleagues18,19 provided a new approach to quantify the proximity principle and formulate its multistability. They found that proximity-based grouping solely depends on interdot distances, not the angle between dot strips or the overall configuration, a finding they termed the Pure Distance Law. They showed that grouping by proximity in dot lattices is a probabilistic competition among potential percepts, where the probability of dots being grouped in a particular direction decreases exponentially as the interdot distances increase18,19. Subsequent research has shown that interdot distances are not the only spatial factor influencing proximity-based grouping in dot lattices; orientation biases also play a role. Several studies13,14,20–22 have demonstrated different patterns of preference for vertical, horizontal, and oblique orientations in dot lattices. In particular, vertical grouping has been frequently observed to be more prevalent than other orientations. The effect of vertical bias on proximity grouping is illustrated in Fig. 1D, where with the same vertical and horizontal interdot distances, it is more probable to perceive the lattice as a group of dots in columns rather than rows. In addition to spatial context, temporal (sequential) context has also been found to influence proximity grouping15. Wertheimer16 observed that when interdot distances were gradually changed in a sequence of stimuli, subjects delayed their perceptual switching from one grouping to another, even when the proximity principle suggested a different grouping. He described this phenomenon as the “factor of objective set”15,16. Decades later, Gepshtein and Kubovy23 empirically tested the effect of temporal context on proximity grouping. In their experiment, a rectangular lattice was followed immediately by a hexagonal lattice in each trial. Their results demonstrated a hysteresis effect, where participants were more likely to perceive the hexagonal lattice in a similar orientation to the preceding rectangular lattice. They also revealed an adaptation effect, in which stronger perceptual evidence along a particular direction led to a decrease in the probability of perceiving the same direction in the following lattice. Subsequent research has used a similar paradigm with slight modifications to study hysteresis and adaptation effects in dot lattices from various perspectives and by employing different computational methods: Schwiedrzik et al.24 studied the neural correlates of these effects using fMRI data and a Bayesian model; Van Geert et al.25 investigated them at an individual rather than group level using a Bayesian hierarchical model comparison approach; and Van Geert et al.26 conducted a simulation study using a Bayesian observer model. Despite these advances, a significant gap remains in the literature. Previous studies on temporal context in perceptual grouping, such as those by Gepshtein and Kubovy23 and their successors24–26, have investigated the influence of a primary stimulus on a secondary stimulus of a different type. However, none of these studies, have analyzed the influence of preceding stimuli in a parametrically varied sequence of identical stimuli, as initially described by Wertheimer16. Consequently, even a century after Wertheimer’s initial observations, there is a lack of experimental research on the dynamics of this perceptual switching. A useful approach to investigate dynamics of multistable perceptual switching using modern computational models is to present a sequence of stimuli where a property is gradually varied on a trial-by-trial basis. This approach provides critical data on transition points and time series of behavioral or neurophysiological responses, which are vital for computational analyses. For example, studies on the bistability of the Necker cube have gradually manipulated the degree of stimulus ambiguity by changing the wireframe contrast of the cube’s edges (their thickness or transparency) in a sequence of trials to investigate the switching between the percepts of a “left-oriented” and a “right-oriented” cube. Behavioral and neurophysiological data were then analyzed using models such as artificial neural networks27, wavelet analysis28, and attractor models29. Similarly, Kim and Frank30 studied the bistability of figure-ground organization by presenting sequences in which the size of four dark squares at the corners of a light square was incrementally increased or decreased. This gradual manipulation of stimulus ambiguity allowed them to apply bifurcation analysis using the Lotka-Volterra-Haken (LVH) amplitude equations to model perceptual switching between “four squares” and “light cross” responses. The LVH model conceptualized these two perceptual decisions as the outcome of competition between distinct modes of brain activity, with stimulus ambiguity serving as a control parameter that modulated the stability of each percept. Ascending and descending sequences revealed hysteresis effects, and comparisons of categorical and continuous judgments showed a positive correlation in model parameters, suggesting a unified self-organizing system underlying both response types. To our knowledge, such computational analyses of dynamic perceptual switching have not yet been applied to perceptual grouping, and existing paradigms, such as that of Gepshtein and Kubovy23, do not provide the necessary data for this approach. Furthermore, the effect of orientation biases on the analysis of temporal context has been largely overlooked. While one study by Van Geert et al.25 did examine the relationship between the magnitude of orientation bias and hysteresis effects, they did not investigate how the direction of orientation bias, in terms of absolute angle in coordinate plane, could affect their hysteresis magnitude. Given the significant influence of orientation biases on perceptual grouping in dot lattices, it is crucial to explore how this spatial context potentially affects the magnitude of the hysteresis effect. The current study addresses key research gaps by investigating the effect of temporal sequential context on proximity-based grouping in rectangular dot lattices, while also considering orientation biases. The proposed method employed rectangular dot lattices presented in ascending and descending sequences, with interdot distances along vector b were gradually increased and decreased, respectively, on a trial-by-trial basis, while distances along vector a remained fixed. This design induced and captured dynamic switching between bistable grouping percepts. To examine the effect of orientation biases, stimuli were presented along two axes (cardinal and oblique) and in two orientations (baseline and perpendicular). This allowed for the specific analysis of perceptual switching between horizontal and vertical responses in the cardinal axis, and between 45° and 135° responses in the oblique axis. The collected data were analyzed to fit psychometric functions and calculate key metrics, including the point of subjective equality (PSE) and hysteresis magnitude. To further examine dynamic bistability, the Lotka-Volterra-Haken (LVH) amplitude equations model was fitted to each axis using a bifurcation analysis approach. The LVH model formalizes these dynamics by providing bifurcation charts that illustrate the observed pattern of perceptual switching as dynamic transitions between fixed points, while also translating behavioral observations into interpretable parameters reflecting the strength of mutual inhibition and the preference for one grouping over another. The PSE, hysteresis, and model parameters were subsequently analyzed using linear mixed-effects models. Methods Participants A 2 × 2 repeated-measures design was employed to investigate the effects of stimulus axis (cardinal and oblique axes) and stimulus orientation (baseline and perpendicular orientations). Power analysis using G*Power 3.131 determined that a sample size of 34 participants would be sufficient to detect a medium effect size (f = 0.25, η2 = 0.06) with power of 0.80 and Type I error (alpha) of 0.05. Data collection was performed on 34 current or former university students (17 female, 17 male), aged 19.4 to 41.2 years (M = 30.5, SD = 6.1). To control for potential lateralization effects on orientation processing, only right-handed subjects were included. Handedness was assessed using the Edinburgh Handedness Inventory32. All subjects had normal or corrected vision and no history of neurodevelopmental or cognitive disorders. Informed consent was obtained from all subjects prior to the study. Experimental procedures adhered to the Declaration of Helsinki and were approved by Research Ethics Committees of Shahid Beheshti University (IR.SBU. REC.1402.016, 2022). Stimuli design To investigate the bistability of proximity-based grouping, we used rectangular dot lattices as visual stimuli. Four stimulus sets were created (Fig. 2A,B): two for the cardinal axis and two for the oblique axis. Each axis included a baseline orientation (original lattice) and a perpendicular orientation (rotated by 90°). In the cardinal axis (Fig. 2A), the basic vectors of the rectangular lattice (a, b) were aligned with 0° and 90° orientations; in the oblique axis (Fig. 2B), they were aligned with 45° and 135° orientations. Across all sets, interdot distances along vector a were fixed, while interdot distances along vector b varied across trials.Fig. 2Stimulus and procedure. (A) Stimulus sets in cardinal axis are presented for two stimulus orientations. Each set comprises 13 stimuli, with the first, seventh, and thirteenth displayed as samples. For each set, the angles of fixed and varying interdot distances are illustrated. (B) Stimulus sets in oblique axis are presented for two stimulus orientations. (C) The procedure for each experimental trial is displayed. Stimulus and procedure. (A) Stimulus sets in cardinal axis are presented for two stimulus orientations. Each set comprises 13 stimuli, with the first, seventh, and thirteenth displayed as samples. For each set, the angles of fixed and varying interdot distances are illustrated. (B) Stimulus sets in oblique axis are presented for two stimulus orientations. (C) The procedure for each experimental trial is displayed. The stimuli were regular arrays of filled black dots on a white background. The dots had a radius of 5 pixels, equivalent to 0.125 degrees of visual angle (dva) at a 60 cm viewing distance. The fixed center-to-center distance between dots along vector a was 60 pixels (1.42 dva). To minimize the influence of the reference frame18, lattices were placed within a circular aperture with radius of 375 pixels (8.90 dva), surrounded by a black rectangle. Edge effects were mitigated by applying a Gaussian filter33 to gradually fade the outer dots within an annulus with inner and outer radii of 4.15 and 8.90 dva, respectively. By keeping interdot distances along a constant and varying distances along b, we generated thirteen unique stimuli defined by AR =|b|/|a|= {2.44–1, 1.77–1, 1.42–1, 1.23–1, 1.12–1, 1.05–1, 1.00, 1.05, 1.12, 1.23, 1.42, 1.77, 2.44}. The AR values were selected such that the grouping direction would be obvious at both ends of the AR range, as previous research17,19 indicates that grouping along a specific vector is unlikely when its interdot distances are more than 1.5 times those of the other vector. The seventh stimulus (AR = 1) corresponded to equal interdot distances along a and b, with AR < 1 and AR > 1 reflecting inverse relations since proximity grouping depends on relative distances10. The selected AR set aimed for a sufficient number of stimuli around AR = 1 to increase the resolution around transition points without significantly increasing the overall number of trials. To achieve this, we employed variable step-sizes, similar to the method used to study Necker cube bistability. For example, Hramov et al.27 and Maksimenko et al.28 both manipulated the contrast of a Necker cube’s inner edges to investigate switching between the "left-oriented" and "right-oriented" percepts. Their control parameter values were {0.1, 0.15, 0.3, 0.4, 0.47, 0.48, 0.49, 0.5, 0.51, 0.52, 0.53, 0.6, 0.7, 0.85, 0.9} and {0.15, 0.25, 0.4, 0.45, 0.5, 0.55, 0.6, 0.75, 0.85}, respectively, with variable step sizes—smaller near the expected switching point (0.5) and larger towards the range boundaries. For the cardinal axis (Fig. 2A), in the baseline orientation, vertical interdot distances (a) were fixed and horizontal distances (b) varied across trials with AR =|b|/|a|. At the first stimulus (AR = 2.44–1, |a|> >|b|), grouping was predominantly horizontal; at seventh (AR = 1, |a| =|b|), vertical and horizontal groupings were equally likely; and at thirteenth (AR = 2.44, |a|< <|b|), grouping was almost exclusively vertical. In the perpendicular orientation, the same principles applied, while horizontal interdot distances (|a|) were fixed and vertical distances (|b|) varied. In both stimulus orientations, the contrasting perceptions at the sequence ends (horizontal vs. vertical) ensured response switching, enabling investigation of transitions in dot lattice perception. For the oblique axis (Fig. 2B), in the baseline orientation, interdot distances were fixed along 135° and varied along 45°; in the perpendicular orientation, 45° was fixed and 135° varied. Here too, the opposing perceptions at the sequence ends (45° vs. 135°) ensured response switching, allowing investigation of transitions between oblique responses. Experimental procedure The experiment included 16 pairs of stimulus presentation sequences, comprising four pairs for each experimental condition: cardinal axis with baseline orientation, cardinal axis with perpendicular orientation, oblique axis with baseline orientation, and oblique axis with perpendicular orientation. The order of the paired-sequences was randomized for each subject to control for order effects. Each paired-sequence included one ascending and one descending sequence. For each subject, half of the paired-sequences were randomly assigned as ascending-first, and the other half as descending-first. In ascending sequences, the aspect ratio (AR) of the stimulus progressed from lower toward higher values on a trial-by-trial basis; in descending sequences, it progressed from higher toward lower values. This paired-sequence design is a standard method for extracting perceptual thresholds in psychophysics34,35, and is adapted from similar studies on the multistability of the Necker cube27–29 and figure-ground discrimination30. To mitigate potential prediction of responses, each sequence began at a randomized initial point near its starting end35,36 (50% from trial 1, 40% from trial 2, 10% from trial 3). The sequence continued until a stable response switch occurred, which was operationally defined as two consecutive trials where the subject’s response differed from their initial response for that sequence35,36. This criterion ensures a robust measurement of the transition point by minimizing the influence of mistakes and uncertainties. The sequence was terminated immediately after a switch was registered. A schematic overview of a single trial of the experiment is illustrated in Fig. 2C. Each trial began with a fixation point displayed for a random duration between 1200 and 1500 ms. Then, the stimulus was presented for 300 ms, followed by a blank screen for 300 ms. Next, the response screen was displayed until the subject’s response. After the subject selected a response using a computer mouse, a 1000-ms blank screen served as the intertrial interval before the next trial began. Participants were seated 60 cm from the screen. They were instructed to fixate on the center of the screen during stimulus presentation and report the initial perceived grouping direction of the dots. The response screen, as shown in Fig. 2C, consisted of two white circles on a black background, each crossed by a diagonal line corresponding to the orientation of the dots alignment along vectors a and b of the lattice. Thus, response options were {0°, 90°} for the cardinal axis and {45°, 135°} for the oblique axis. To control for potential preferences for the left or right response choices, the location of response options (a and b) in the response screen were counterbalanced between paired-sequences but kept fixed within each. Before data collection, a training session was conducted to familiarize subjects with the stimuli and experimental instructions. In the training session, four paired-sequences were presented to each subject in random order, including one pairs for each experimental condition: cardinal-baseline, cardinal-perpendicular, oblique-baseline, and oblique-perpendicular. All experimental steps, including stimuli generation, task design, and data collection, were carried out on an ACER Aspire 5750G laptop with an Intel Core i5-2450M processor, 8GB of RAM, and a 15.6-inch HD LED LCD (1024 × 768 pixels) with a reported static contrast ratio of 196:1. The display brightness was set to its maximum level (100%) during data collection. Experiment design and implementation were accomplished using MATLAB software (Version R2016b) with the Psychtoolbox-3 extension37,38. Data preparation and analysis Data were analyzed for the four experimental conditions, which resulted from a 2 (stimulus axis: cardinal, oblique) × 2 (stimulus orientations: baseline, perpendicular) design. For each presentation sequence, a perceptual switch was defined as a change in response for two consecutive trials. The transition point was then calculated as the stimulus number halfway between the final trial of the initial percept and the first trial of the new, stable percept (e.g., for a response sequence of {b, b, b, b, b, b, a, a}, the switch between stimulus 6 and 7 defines the transition point as 6.5). Because each experimental condition included four paired-sequences, this procedure yielded four ascending and four descending transition points per participant for each condition. These were subsequently averaged to compute the mean descending \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{1}^{ * }$$\end{document} and ascending \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{2}^{ * }$$\end{document} transition points. These raw transition points, recorded on a scale of stimulus number from 1 to 13, correspond to an aspect ratio (AR) interval of 2.44–1 to 2.44. To perform a preliminary data analysis and evaluate the model fit, a psychometric function34,35 was extracted for each experimental condition averaged over the participants. First, the number of “yes” responses was determined as a function of stimulus intensity, defined as the aspect ratio (AR =|b|/|a|). In the cardinal axis, “yes” responses corresponded to vertical grouping in the baseline orientation and horizontal grouping in the perpendicular orientation; in the oblique axis, they corresponded to 135° and 45° groupings, respectively. The data points for “yes” responses were the mean of data in ascending and descending sequences, as suggested in classical methods34,35. Next, a logistic function was fitted to the mean data points to obtain the psychometric function curve. Additionally, this procedure was performed separately for the ascending and descending data in order to demonstrate the hysteresis effect. All psychometric functions were calculated using the Palamedes toolbox in MATLAB. To extract the variables for statistical analysis, the point of subjective equality (PSE) was calculated using equation (1) as the mean of the ascending and descending transition points. Additionally, the hysteresis was calculated using equation (2) as the difference between the ascending and descending transition points. Equation (3) was used to scale the raw transition points by stimulus number to the [0-1] interval:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$PSE = ~\frac{{\alpha _{2}^{*} + \alpha _{1}^{*} }}{2}$$\end{document}2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Hysteresis = ~\alpha _{2}^{*} - \alpha _{1}^{*}$$\end{document}3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha _{1} = ~\frac{{\alpha _{1}^{*} - 1}}{{13 - 1}},~~~~~~\alpha _{2} = ~\frac{{\alpha _{2}^{*} - 1}}{{13 - 1}}$$\end{document} Data modeling The Lotka-Volterra-Haken (LVH) model, described in detail by Frank and Kim30,39–41 is a bifurcation analysis framework within dynamical systems theory. It originates from the classical Lotka-Volterra model in ecology, which was initially developed to describe predator–prey dynamics and later extended to competing species vying for common resources42,43. In this context, the model characterizes how each species’ population changes over time as a function of its intrinsic growth rate, self-inhibition (competition within species due to limited resources), and cross-inhibition (competition between species for shared resources). When cross-inhibition dominates self-inhibition, the model predicts competitive exclusion, where only one species survives. This principle of nonlinear competition provides a versatile framework for explaining how mutually exclusive states emerge and stabilize in diverse systems. This concept of competition was generalized beyond ecology by Hermann Haken44 and others within the theory of self-organization, or synergetics. They reformulated the Lotka-Volterra equations as coupled amplitude equations (order parameter equations) that govern the emergence of spatio-temporal patterns in complex systems. While the classical Lotka-Volterra framework typically uses quadratic nonlinearities (d = 2), Haken’s model employs cubic nonlinearities (d = 3); despite this difference, the two forms are mathematically equivalent through a variable transformation. The LVH model conceptualizes competition between perceptual alternatives as the competitive interaction of distinct modes of brain activity, each represented by a time-dependent amplitude reflecting its activation level39–41. The dynamics of these amplitudes are governed by coupled equations with strong mutual inhibition, ensuring that only one percept dominates at a time in a “winner-takes-all” manner, a defining feature of mutually exclusive behavioral patterns. Gradual changes in stimulus properties act as a control parameter within model, altering the stability of each percept’s attractor. At a critical threshold—a bifurcation—the currently dominant percept becomes unstable, triggering a transition to the alternative percept. According to the LVH model, in a two-choice perceptual task, the amplitudes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} represent the activity level of the neural populations associated with each response. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}>0$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}=0$$\end{document}, the first response dominates; if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}=0$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}>0$$\end{document}, the second response dominates. The amplitudes are dynamic, reflecting the time-varying nature of brain activity, and their evolution is computed using the coupled amplitude Equations.30,41:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ {\begin{array}{*{20}c} {\frac{d}{{dt}}A_{1} = \lambda _{1} A_{1} - A_{1}^{d} - gA_{2}^{{d - 1}} A_{1} } \\ {\frac{d}{{dt}}A_{2} = \lambda _{2} A_{2} - A_{2}^{d} - gA_{1}^{{d - 1}} A_{2} } \\ \end{array} } \right.$$\end{document} Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d}{dt}{A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d}{dt}{A}_{2}$$\end{document} are the time derivatives of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}, respectively; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{2}$$\end{document} are the growth parameters of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}, respectively, representing the stimulus evidence or bottom-up drive for each percept; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} is the mutual inhibition parameter that reflects the strength of the competition between the neural populations; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d$$\end{document} is the order of the model’s nonlinearity and equals 3 in the LVH model. In this equation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{1}{A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{2}{A}_{2}$$\end{document} capture the linear dynamic changes of the amplitudes over time. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}^{d}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}^{d}$$\end{document} represent saturation or self-inhibition, preventing unbounded neural activity—a biologically plausible constraint. The interaction terms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g{A}_{2}^{d-1}{A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g{A}_{1}^{d-1}{A}_{2}$$\end{document} denote cross-inhibition between the neural populations. In the current study, the response options corresponded to the grouping of dots along vectors a and b of the rectangular lattice. Stimuli were presented in ascending and descending sequences, with the scaled stimulus index (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}), ranging from 0 to 1, serving as the task’s control parameter. Gradual increases or decreases in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} produced corresponding changes in interdot distances, inducing response switches between the two possible percepts. Assuming \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{1}=A\left(1-\alpha \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{2}=B\alpha$$\end{document}, an increase in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} over time led to a linear decrease in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{1}$$\end{document} and a linear increase in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\lambda }_{2}$$\end{document}. The preference parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma =\frac{A}{B}$$\end{document}) quantified the relative tendency toward the response associated with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} versus \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma >1$$\end{document} indicating a preference for the first response and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma <1$$\end{document} indicating a preference for the second. Using the scaled descending (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{1}$$\end{document}) and ascending (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{2}$$\end{document}) transition points, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} were estimated as follows30:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma = \sqrt {\frac{{\alpha _{1} \alpha _{2} }}{{\left( {1 - \alpha _{1} } \right)\left( {1 - \alpha _{2} } \right)}}}$$\end{document}6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g = \sqrt {\left( {\frac{{\alpha _{1} }}{{\alpha _{2} }}} \right)\left( {\frac{{1 - \alpha _{1} }}{{1 - \alpha _{2} }}} \right)}$$\end{document} We fitted the data to the LVH model at two levels: the group level, to demonstrate the model’s performance using descriptive charts, and the individual level, to extract model parameters for statistical comparisons. Perceptual switching for each stimulus axis occurred independently; therefore, separate LVH models were fit for the cardinal and oblique axes. Within each axis, data from baseline and perpendicular orientations were combined. In the cardinal axis, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} represented the amplitudes of brain activity associated with vertical (90°) and horizontal (0°) groupings, respectively, making \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} the relative preference for vertical over horizontal responses. In the oblique axis, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} corresponded to 135° and 45° groupings, respectively, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} reflecting the preference for 135° over 45° responses. To further explore the LVH model results for the cardinal and oblique axes, fixed points were calculated and bifurcation curves drawn. Stable fixed points were computed using the Matcont Toolbox in MATLAB, with the control parameter set to its boundary values (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 1). Bifurcation curves were obtained via numerical simulations using MATLAB’s ode45 function. These results were used to plot the dynamic trajectories of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} amplitudes over time and as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} for both group and individual levels. Intersection points of the amplitude trajectories were calculated to represent simulated transition points, which were then compared with observed values to evaluate model accuracy. Additionally, to evaluate the predictive power of the fitted models, simulated responses were compared with the empirical data using psychometric functions. The simulated responses were determined by comparing the corresponding amplitudes at individual level: the first response was assigned if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}>{A}_{2}$$\end{document}, and the second response if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}<{A}_{2}$$\end{document}. Statistical analysis For statistical analysis, we employed linear mixed-effects models (LMMs) in the R environment45 using the lme4 package46. Mixed-effects models47,48 incorporate subject-by-subject variations in model parameters (random effects) alongside fixed effects. They offer advantages over traditional methods (e.g., repeated measures ANOVA and multiple regression) by not requiring the independence assumption, accounting for hierarchical data structures, and allowing the simultaneous analysis of multiple fixed and random effects49. To find the best-fitting models given the data and to estimate the effect of experimental conditions on the responses, linear mixed models were utilized. The dependent variables were the point of subjective equality (PSE), hysteresis magnitude, preference parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}), and mutual inhibition parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}). In the analysis of PSE and hysteresis, fixed effects of Axis (oblique vs. cardinal axis) and Orientation (perpendicular vs. baseline orientation) were included. To form the mixed models, all possible combinations of these fixed effects and their interaction were considered, resulting in five models: Axis × Orientation, Axis + Orientation, Axis, Orientation, and Null models. Random effects were included to account for subject-by-subject variations in the intercept and slope of Axis and Orientation. In the analysis of the LVH model parameters (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}), the fixed effect of Axis was considered, while random effects accounted for subject-by-subject variations in the intercept and slope of Axis. This analysis included two models: Axis and Null models. To select the best-fitting linear mixed model for each response variable, a model comparison and selection process was employed. This process utilized the corrected Akaike Information Criterion (AICc) as implemented in the AICcmodavg package50. While the AIC provides a balanced evaluation of model fit and complexity, its corrected version, AICc, addresses potential bias associated with smaller sample sizes51. To compare the models, several criteria were calculated and reported: number of model parameters (K), AICc difference from the top model (ΔAICc), AICc weight, and evidence ratio (Bayes factor). If a single model was clearly superior, inference was drawn from that model. Otherwise, if multiple models were competitive, multi-model inference was employed51. A top model was selected as the KL best model if it had a strong evidence ratio52 (> 10) and the remaining models had considerably less empirical support51 (ΔAICc > 4). If not, a combined model was constructed using model averaging on the competitive models, implemented using the MuMIn package53. The AICc effectively compares and selects models, but it does not assess absolute model fit. To evaluate this, marginal R2 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{marg.}^{2}$$\end{document}) and conditional R2 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{cond.}^{2}$$\end{document}), calculated using MuMIn package, measure the variance explained by fixed and both fixed and random effects, respectively. To identify significant effects, 95% confidence intervals (CIs) were calculated and reported for fixed effect estimates, rather than null-hypothesis testing14,54. An effect was considered significant at the p < 0.05 level if its confidence interval did not include zero. To further explore the results, pairwise post-hoc comparisons of estimated marginal means (EMMs) with Tukey’s adjustment were conducted using the emmeans package55. Pairwise contrasts were considered significant if their 95% confidence intervals (CIs) did not span zero. Results Psychometric functions Figure 3A shows the proportion of vertical grouping responses for stimuli in the cardinal axis for the baseline orientation. Mean data points, averaged across ascending and descending sequences, ranged from 0 to 100%. The first stimulus (AR = 2.44–1) was always perceived horizontally, yielding zero vertical responses. As AR increased, vertical responses gradually rose, with the final stimulus (AR = 2.44) consistently perceived vertically. The fitted psychometric function shows a strong goodness-of-fit (R2 = 0.9985). The 50% threshold (AR = 0.9028) indicates an early switch from horizontal to vertical responses relative to the equilibrium point (seventh stimulus, AR = 1). Comparison of ascending and descending curves reveals hysteresis: the switch from horizontal to vertical in ascending sequences occurs later than the reverse in descending sequences, forming a hysteresis loop.Fig. 3Psychometric functions fitted to the data across four experimental conditions. (A) Proportion of vertical responses in the cardinal axis with baseline orientation is displayed as a function of aspect ratio (AR). Green points show the mean of ascending and descending sequences, with the solid black curve fitted to the mean. Separate curves were fitted to ascending (red dashed) and descending (blue dash-dot) sequences. All horizontal axes are logarithmic. The 50% perceptual threshold and R2 for the mean curve are provided. (B–D) Psychometric functions for the remaining conditions are displayed: (B) horizontal responses, cardinal-perpendicular, (C) 135° responses, oblique-baseline, and (D) 45° responses, oblique-perpendicular. Psychometric functions fitted to the data across four experimental conditions. (A) Proportion of vertical responses in the cardinal axis with baseline orientation is displayed as a function of aspect ratio (AR). Green points show the mean of ascending and descending sequences, with the solid black curve fitted to the mean. Separate curves were fitted to ascending (red dashed) and descending (blue dash-dot) sequences. All horizontal axes are logarithmic. The 50% perceptual threshold and R2 for the mean curve are provided. (B–D) Psychometric functions for the remaining conditions are displayed: (B) horizontal responses, cardinal-perpendicular, (C) 135° responses, oblique-baseline, and (D) 45° responses, oblique-perpendicular. Figure 3B displays the proportion of horizontal responses for the perpendicular orientation in the cardinal axis. The first stimulus was always perceived vertically, giving zero horizontal responses. As AR increased, horizontal responses rose, reaching 100% at the final stimulus. The fitted curve exhibits strong fit (R2 = 0.9969), with a 50% threshold (AR = 1.1693) indicating a late switch from vertical to horizontal responses. Hysteresis is evident, minimal for AR ≤ 1 but increasing for AR > 1. Figure 3C and D show psychometric functions for 135° and 45° responses in the oblique axis for the baseline and perpendicular orientations, respectively. In Fig. 3C, the first stimulus (AR = 2.44–1) was always perceived along 45°, producing zero 135° responses. The 135° responses increased with AR, reaching 100% at the final stimulus (AR = 2.44). In Fig. 3D, the first stimulus was always perceived along 135°, yielding zero 45° responses; 45° responses increased with AR to 100% at the final stimulus. Goodness-of-fit was high for both plots (R2 = 0.9934 baseline; R2 = 0.9921 perpendicular). The 50% thresholds (AR = 1.0056 baseline; AR = 1.0253 perpendicular) were near the equilibrium point (AR = 1), indicating slightly later switching from 135° to 45° responses. Ascending and descending curves formed hysteresis loops, which were wider for the baseline than for the perpendicular orientation. Point of subjective equality (PSE) Figure 4A shows the PSE variation across stimulus axes and orientations in terms of stimulus number and aspect ratio (AR). A notable difference was observed between baseline and perpendicular orientations in the cardinal axis, while the oblique axis showed little difference.Fig. 4Point of subjective equality (PSE). (A) Error bars indicate the standard error (SE) across stimulus axes and orientations. PSE values are shown in terms of stimulus number and aspect ratio (AR). (B) Pairwise contrasts of PSE between experimental conditions are presented, with estimated values ± 95% CIs. Point of subjective equality (PSE). (A) Error bars indicate the standard error (SE) across stimulus axes and orientations. PSE values are shown in terms of stimulus number and aspect ratio (AR). (B) Pairwise contrasts of PSE between experimental conditions are presented, with estimated values ± 95% CIs. In the cardinal axis with baseline orientation, the transition from horizontal to vertical grouping occurred at point 5.54, much earlier than the equilibrium point (seventh stimulus, AR = 1). In the perpendicular orientation, the initial vertical perception persisted, with a remarkably late transition to horizontal grouping at point 8.89. These transition points indicate a persistent preference for vertical over horizontal grouping, suggesting a vertical bias. In the oblique axis, baseline orientation showed a transition from 45° to 135° grouping at point 7.09 near the equilibrium point, and perpendicular orientation showed a transition from 135° to 45° at point 7.34, with a small delay. These results indicate only a slight bias toward 135° grouping. To analyze the effect of experimental conditions on PSE, five linear mixed-effects models were fitted to the data. The results are detailed in Supplementary Table S1 online, with models sorted by ΔAICc values. The first model (Axis × Orientation) outperformed the others and was selected for inference, as other models had ΔAICc > 4 and evidence ratios > 10. The selected model’s marginal and conditional R2 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{marg.}^{2}$$\end{document} = 0.566, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{cond.}^{2}$$\end{document} = 0.619) indicated a large amount of variance explained by fixed and random effects. The estimated parameters of the chosen model are presented in Table 1 (panel A). The effects of Axis, Orientation, and their interaction (Axis:Orientation) were significant, as their 95% CIs did not include zero.Table 1Results of model coefficients for PSE (panel A), hysteresis (panel B), preference parameter (panel C), and mutual inhibition parameter (panel D). For each variable, the estimated coefficient is presented, along with the standard error (SE) and 95% lower and upper confidence intervals (CIs).EffectEstimateSELower CIUpper CIPanel A: PSE Intercept5.5440.1685.2125.876 Axis1.5480.2401.0732.023 Orientation3.3460.2582.8363.855 Axis:Orientation− 3.0970.335− 3.760− 2.434Panel B: Hysteresis Intercept0.8770.3130.2581.496 Axis0.0130.124− 0.3910.472 Orientation− 0.1940.190− 0.6120.065 Axis:Orientation− 0.0350.139− 0.9300.209Panel C: Preference parameter (γ) Intercept1.9860.1081.7722.201 Axis− 0.9310.148− 1.230− 0.631Panel D: Mutual inhibition parameter (g) Intercept1.0590.0121.0361.082 Axis− 0.0530.015− 0.084− 0.022 Table 1Results of model coefficients for PSE (panel A), hysteresis (panel B), preference parameter (panel C), and mutual inhibition parameter (panel D). For each variable, the estimated coefficient is presented, along with the standard error (SE) and 95% lower and upper confidence intervals (CIs).EffectEstimateSELower CIUpper CIPanel A: PSE Intercept5.5440.1685.2125.876 Axis1.5480.2401.0732.023 Orientation3.3460.2582.8363.855 Axis:Orientation− 3.0970.335− 3.760− 2.434Panel B: Hysteresis Intercept0.8770.3130.2581.496 Axis0.0130.124− 0.3910.472 Orientation− 0.1940.190− 0.6120.065 Axis:Orientation− 0.0350.139− 0.9300.209Panel C: Preference parameter (γ) Intercept1.9860.1081.7722.201 Axis− 0.9310.148− 1.230− 0.631Panel D: Mutual inhibition parameter (g) Intercept1.0590.0121.0361.082 Axis− 0.0530.015− 0.084− 0.022 Results of model coefficients for PSE (panel A), hysteresis (panel B), preference parameter (panel C), and mutual inhibition parameter (panel D). For each variable, the estimated coefficient is presented, along with the standard error (SE) and 95% lower and upper confidence intervals (CIs). To demonstrate how these significant effects affected the PSE, post-hoc pairwise tests were conducted and their results are illustrated in Fig. 4B. The results revealed significant differences between the cardinal and oblique axes in both stimulus orientations, as their 95% CIs excluded zero. Notably, a significant difference between baseline and perpendicular orientations was found in the cardinal axis, but not in the oblique axis. These results indicate a significant orientation bias toward vertical responses in the cardinal axis, while no significant bias was observed in the oblique axis. Transition points and hysteresis Figure 5A shows error bars for transition points in ascending and descending sequences. Due to the preference for vertical grouping in the cardinal axis, the transition from 0° to 90° grouping in the baseline orientation occurred early (between stimuli 5 and 6), while the transition from 90° to 0° in the perpendicular orientation occurred late (around stimulus 9). In the oblique axis, for both orientations, transition points were distributed around the equilibrium point (seventh stimulus, AR = 1), between stimuli 6 and 8. Ascending transition points were consistently higher than descending ones across conditions, indicating a hysteresis effect, consistent with the curves in Fig. 3.Fig. 5Transition points and hysteresis. (A) Error bars show the standard error for transition points across stimulus axes and orientations in ascending and descending sequences. (B) Error bars display the standard error for hysteresis across axes and orientations. (C) Pairwise contrasts of hysteresis between experimental conditions are presented, with estimated values ± 95% CIs. (D) Correlation between hysteresis values in cardinal and oblique axes is shown, with estimated correlation coefficient (r) and ± 95% CI. Transition points and hysteresis. (A) Error bars show the standard error for transition points across stimulus axes and orientations in ascending and descending sequences. (B) Error bars display the standard error for hysteresis across axes and orientations. (C) Pairwise contrasts of hysteresis between experimental conditions are presented, with estimated values ± 95% CIs. (D) Correlation between hysteresis values in cardinal and oblique axes is shown, with estimated correlation coefficient (r) and ± 95% CI. Figure 5B shows error bars for hysteresis magnitude, calculated as the difference between ascending and descending transition points. Hysteresis was positive in all conditions, slightly higher in the baseline orientation, especially in the oblique axis. This pattern mirrors the width of hysteresis loops in Fig. 3. To test the effect of factors on hysteresis, five linear mixed-effects models were fitted (see Supplementary Table S2 online). The first model (Orientation) had negligible variance explained by fixed effects (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{marg.}^{2}$$\end{document} = 0.007) but a large portion of variance explained by both fixed and random effects (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{cond.}^{2}$$\end{document} = 0.777). Since all five models were competitive (ΔAICc < 4; evidence ratio < 10), they were selected for model averaging. Table 1 (panel B) shows the averaged results. The intercept was significant, with a 95% CI ranging from 0.258 to 1.469. This result highlighted the significant existence of a hysteresis effect overall. Other fixed effects were not deemed significant (95% CIs encompassed zero), suggesting no effect of experimental conditions on hysteresis. Post-hoc pairwise tests (Fig. 5C) likewise showed no significant differences across conditions. Figure 5D plots individual participants’ hysteresis values between stimulus axes, with a fitted regression line. A strong positive correlation (r = 0.806, 95% CI = [0.643, 0.899]) was observed between hysteresis in the cardinal and oblique axes. Lotka-Volterra-Haken (LVH) model Figure 6A shows numerically simulated amplitude trajectories of the LVH model in the cardinal axis at both individual and group levels. The plot illustrates dynamic transitions between stable fixed points over time, reflecting bifurcations induced by changes in the control parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( \alpha \right)$$\end{document}. At the group level, when horizontal (0°) grouping initially dominates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}\ll {A}_{2}$$\end{document}), the trajectories converge to the first stable point (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} = 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 1). As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} increases, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} rises and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} decreases, leading to their intersection. After this point, vertical (90°) grouping dominates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}\gg {A}_{2}$$\end{document}), and the trajectories converge to the second stable point (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} = 1.3326, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 0). The higher value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} compared to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} reflects the relative preference for vertical over horizontal responses, as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}/{A}_{2}$$\end{document} ratio is proportional to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{\gamma }$$\end{document}. These trajectories can also be interpreted in reverse, when decreasing α induces a transition from the second stable point (90° grouping) back to the first (0° grouping). At the individual level, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} trajectories follow the same overall pattern as the group, but differences in participants’ model parameters result in distinct fixed points and amplitudes. Individual \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} values match the group-level results because, in the LVH model, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} serves as a base amplitude and calculations depend on the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}/{A}_{2}$$\end{document} ratio, so it is not affected by individual model parameters.Fig. 6Simulation results of the Lotka-Volterra-Haken (LVH) model. (A) Bifurcation curve for the cardinal axis as a function of time. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} represent dynamic amplitudes of brain activity patterns associated with vertical (90°) and horizontal (0°) groupings, respectively. Blue dotted lines show individual participants’ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}, while blue solid and red dash-dot lines show group-level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}, respectively. Note that individual-level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} results match group-level results. With \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A$$\end{document} = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} = 1.7758, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B$$\end{document} = 1.0, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} = 1.0290, group-level fixed points were (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 1) and (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 1.3326, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}= 0). (B) Bifurcation curve for the cardinal axis as a function of control parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} trajectories follow the definitions in panel A. The intersection point of group-level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} is indicated. (C) Psychometric functions for the cardinal axis as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}. Red and blue lines represent the proportions of 0° and 90° responses, respectively, for actual (solid) and simulated (dash-dot) data. Simulated data were derived from the integration of individual-level model results. Intersection points are marked as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{Act}$$\end{document} for actual and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{Sim}$$\end{document} for simulated curves. (D) Bifurcation curve for the oblique axis as a function of time. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} denote activity amplitudes for 135° and 45° groupings, respectively. With \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A$$\end{document} = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} = 1.0432, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B$$\end{document} = 1.0, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} = 1.0024, group-level fixed points were (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 1) and (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 1.0214, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}= 0). (E) Bifurcation curve for the oblique axis as a function of control parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}). (F) Psychometric functions for the oblique axis as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}. Red and blue lines represent the proportions of 45° and 135° responses, respectively, for actual (solid) and simulated (dash-dot) data. Simulation results of the Lotka-Volterra-Haken (LVH) model. (A) Bifurcation curve for the cardinal axis as a function of time. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} represent dynamic amplitudes of brain activity patterns associated with vertical (90°) and horizontal (0°) groupings, respectively. Blue dotted lines show individual participants’ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}, while blue solid and red dash-dot lines show group-level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}, respectively. Note that individual-level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} results match group-level results. With \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A$$\end{document} = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} = 1.7758, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B$$\end{document} = 1.0, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} = 1.0290, group-level fixed points were (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 1) and (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 1.3326, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}= 0). (B) Bifurcation curve for the cardinal axis as a function of control parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} trajectories follow the definitions in panel A. The intersection point of group-level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} is indicated. (C) Psychometric functions for the cardinal axis as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}. Red and blue lines represent the proportions of 0° and 90° responses, respectively, for actual (solid) and simulated (dash-dot) data. Simulated data were derived from the integration of individual-level model results. Intersection points are marked as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{Act}$$\end{document} for actual and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{Sim}$$\end{document} for simulated curves. (D) Bifurcation curve for the oblique axis as a function of time. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} denote activity amplitudes for 135° and 45° groupings, respectively. With \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A$$\end{document} = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} = 1.0432, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B$$\end{document} = 1.0, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} = 1.0024, group-level fixed points were (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 1) and (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document}= 1.0214, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document}= 0). (E) Bifurcation curve for the oblique axis as a function of control parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}). (F) Psychometric functions for the oblique axis as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}. Red and blue lines represent the proportions of 45° and 135° responses, respectively, for actual (solid) and simulated (dash-dot) data. Figure 6B displays amplitude trajectories in the cardinal axis at individual and group levels as a function of the control parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}). As in Fig. 6A, changes in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} induce transitions between stable points: 0° responses dominate before the intersection point and 90° responses dominate after. The early intersection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0.3495), well before \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0.5, reflects a strong vertical grouping preference, consistent with previous results. The observed descending and ascending transition points were \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{1}$$\end{document} = 0.3370 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{2}$$\end{document} = 0.3842, with a mean of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0.3606—only 3.08% different from the simulated value (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0.3495)—demonstrating that the LVH model closely approximates the cardinal-axis transition point. Figure 6C presents the proportions of 0° and 90° responses across \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} values in the cardinal axis. Actual and simulated data points are shown, with fitted psychometric curves. Simulated data were derived by integrating individual-level model results. The patterns of actual and simulated data are similar: 0° responses decrease and 90° responses increase with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}. The curves closely align, with minor differences arising because the LVH model does not capture trial-by-trial variability. The simulated intersection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{Sim} = 0.3568$$\end{document}) closely matches the actual intersection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{Act} = 0.3611$$\end{document}), differing by only 1.2%, demonstrating strong predictive power of the LVH model in the cardinal axis. Figure 6D shows numerically simulated amplitude trajectories of the LVH model and their bifurcation in the oblique axis at individual and group levels. At the group level, when 45° grouping initially dominates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}\ll {A}_{2}$$\end{document}), trajectories converge to the first stable point (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} = 0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 1). As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( \alpha \right)$$\end{document} increases, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} rises and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} decreases until intersection, after which 135° grouping dominates (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}\gg {A}_{2}$$\end{document}), converging to the second stable point (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} = 1.0214, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{2}$$\end{document} = 0). Individual \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{1}$$\end{document} trajectories follow the same overall pattern, but differences in participants’ parameters yield distinct fixed points. Compared to the cardinal axis, individual deviations are smaller. Figure 6E shows amplitude trajectories in the oblique axis as a function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}. The early intersection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0.4888), just before \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0.5, indicates a slight tendency toward 135° grouping, consistent with previous results. Observed mean transition point was \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} = 0.4895 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{1}$$\end{document} = 0.4617 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{2}$$\end{document} = 0.5172), differing from the simulated value by only 0.13%, demonstrating high model accuracy. Figure 6F presents the proportions of 45° and 135° responses across \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} values in the oblique axis. Patterns of actual and simulated data are similar, with 45° responses decreasing and 135° responses increasing. The curves closely align, with minor differences due to trial-by-trial variability of actual data. The simulated intersection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{Sim} \, = \,0.4927$$\end{document}) closely matches the actual intersection (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha_{Act} = 0.4857$$\end{document}), differing by 1.4%, showing strong predictive power of the LVH model in the oblique axis. Figure 7A plots the preference parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}) for each stimulus axis. Values are higher in the cardinal axis (just below 2) than in the oblique axis (just above 1). These reflect a strong vertical bias and a slight 135° bias, since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma >1$$\end{document} indicates preference for vertical/135° over horizontal/45° responses. Figure 7B shows the mutual inhibition parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}), slightly above 1 in both axes.Fig. 7Parameters of the LVH model. (A) Error bars show the standard error of the preference parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}) across stimulus axes. (B) Error bars indicate the standard error of the mutual inhibition parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}) across axes. (C) Pairwise contrast of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} between experimental conditions, with estimated values ± 95% CIs. (D) Pairwise contrast of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} between experimental conditions, with estimated values ± 95% CIs. Parameters of the LVH model. (A) Error bars show the standard error of the preference parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}) across stimulus axes. (B) Error bars indicate the standard error of the mutual inhibition parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}) across axes. (C) Pairwise contrast of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} between experimental conditions, with estimated values ± 95% CIs. (D) Pairwise contrast of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} between experimental conditions, with estimated values ± 95% CIs. To evaluate the impact of stimulus axis on the preference and mutual inhibition parameters, two linear mixed models were fitted for each parameter (Supplementary Table S3 online). Based on ΔAICc and evidence ratio, the Axis model outperformed the Null model for both parameters and was selected for inference. Estimated coefficients are presented in Table 1 (panels C-D). Model coefficients indicate that both preference (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}) and mutual inhibition (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}) parameters were significantly influenced by Axis, as their 95% CIs excluded zero. Pairwise post-hoc tests (Fig. 7C,D) confirm significantly higher values for both parameters in the cardinal axis compared to the oblique axis. Discussion In this study, we investigated how temporal and spatial contexts influence multistable perception of proximity-based grouping in rectangular dot lattices. Temporal effects were examined by systematic manipulating interdot distances in ascending and descending sequences. Stimuli were also presented along two axes (cardinal and oblique) and two orientations (baseline and perpendicular) to account for orientation biases. Psychometric functions demonstrated perceptual switching and hysteresis loops, and showed excellent fits to the data. Ascending transition points consistently exceeded descending points across experimental conditions, confirming a hysteresis effect. Interestingly, this hysteresis effect was unaffected by stimulus orientation and positively correlated across axes. Analysis of psychometric functions and transition points revealed a strong vertical bias in the cardinal axis and a slight, non-significant bias toward 135° in the oblique axis. Bifurcation curves from numerical simulations of the Lotka-Volterra-Haken (LVH) model demonstrated bistability of proximity-based grouping with good accuracy. Comparisons of simulated and actual data indicated strong predictive power of the fitted models, and model parameters reflected the observed patterns of grouping preference. The following section will discuss these key findings in more detail. Our findings revealed hysteresis in perceptual grouping by the proximity principle in rectangular dot lattices. This experimentally confirms Wertheimer’s observation16 that subjects tend to persist in a perceived grouping pattern, delaying response switching when stimulus properties change. However, Wertheimer provided only a qualitative account of this phenomenon, without experimental quantification. This gap reflected the methodological approach of early Gestalt psychologists, which emphasized phenomenological demonstrations—compelling visual proofs—over quantitative analysis56,57. Hysteresis, common in both animate and inanimate systems58, has been observed across many perceptual domains59. Positive hysteresis, the more common form, acts as a priming effect: the prior perceptual state facilitates perceiving the current stimulus in the same way, creating perceptual inertia despite changes in stimulus properties60,61. In contrast, negative hysteresis arises when neural adaptation from prior exposure inhibits perceiving subsequent stimuli, reducing the likelihood of similar perception41,61. In this study, hysteresis was positive in all experimental conditions. This aligns with prior findings in proximity-based grouping23–26, figure-ground discrimination30,62 and apparent motion63 in the broader field of perceptual organization. While our study confirms hysteresis in proximity-based grouping, consistent with Gepshtein and Kubovy23, our methodology differs significantly. First, we used a parametrically varied sequence of identical stimuli, allowing extraction of psychometric functions, transition points, and bifurcation analysis of perceptual switching dynamics. In contrast, Gepshtein and Kubovy examined how a primary lattice influenced a secondary, different lattice, without providing data required for dynamical analysis. Second, whereas their method controlled orientation with random angles, our design directly investigated response switching and hysteresis at specific orientations, enabling analysis of orientation biases in hysteresis. The results showed that the hysteresis effect emerged in both stimulus axes, despite orientation biases. Unequal preferences for grouping along different orientations did not affect its occurrence or strength. Specifically, although orientation bias caused a significant shift in transition points toward vertical responses, the magnitude and sign of hysteresis—the difference between ascending and descending transition points—remained constant. This indicates that hysteresis in proximity-based grouping is robust against orientation biases, perhaps because it reflects temporal context and relies on a more fundamental mechanism than spatial-context-based orientation bias. This could provide valuable clues for future research on how temporal and spatial contexts interact in shaping perceptual grouping. Our experimental design also controlled for non-perceptual response biases by randomizing the order of paired-sequences (across stimulus axes and orientations), counterbalancing ascending-first and descending-first sequences, and counterbalancing response positions on the screen. Participants were also blind to the study’s goals. While we cannot fully rule out unknown non-perceptual biases, our findings likely reflect a genuine perceptual phenomenon, as such biases would shift both ascending and descending transition points similarly, leaving hysteresis magnitude unchanged. The correlation analysis revealed a significant positive relationship between hysteresis magnitudes in the cardinal and oblique axes, indicating consistency across experimental conditions. This aligns with Kim and Frank30, who observed positive correlations between participants’ hysteresis in different perceptual tasks, and Van Geert et al.25, who demonstrated stable hysteresis patterns across experimental sessions. These observations suggest that hysteresis may be a personal perceptual characteristic, a hypothesis that warrants further investigation. The Lotka-Volterra-Haken (LVH) model30,39–41 posits that perceptual decisions arise from competition between alternative percepts, each associated with distinct neural populations represented by time-dependent amplitudes. Previously, the model has been applied to perceptual switching in tasks such as size perception39–41, speed perception39,40, and figure-ground discrimination30. To our knowledge, this is the first application of the LVH model to perceptual grouping. Our results show that the LVH model successfully explains behavioral observations along both cardinal and oblique axes. Simulated amplitude trajectories captured dynamic perceptual switching between {0°, 90°} and {45°, 135°} responses, reproducing the winner-take-all nature of perceptual dominance. The plotted curves demonstrated that gradual changes in the stimulus property (AR), expressed through the control parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}), can trigger a sudden change in perceptual response, a key feature of bistability. These results revealed that transitions between perceptual states are not arbitrary; they are formal bifurcations, where a stable attractor becomes unstable (a saddle point) as the control parameter crosses a critical threshold. Comparison of actual and simulated psychometric curves further demonstrated the strong predictive power of the model, which accurately estimated transition points across both axes. These findings establish the LVH model as a powerful framework for describing the dynamics of perceptual switching in dot lattices. Importantly, the LVH model moves beyond behavioral description by providing a mechanistic account and linking observed dynamics to neurophysiologically interpretable parameters. While metrics such as the Point of Subjective Equality (PSE) and hysteresis magnitude indicate what happens during switching, the LVH model explains why by decomposing the behavioral metrics into two latent parameters: preference (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}) and mutual inhibition (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}). These cannot be measured directly from behavior but instead reflect underlying system properties that generate observed outcomes. In our data, transition points and PSEs revealed a strong vertical bias in the cardinal axis and a mild 135° bias in the oblique axis, consistent with prior work13,14,20–22. These asymmetries were quantitatively captured by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}, which was just below 2 for the cardinal axis and just above 1 for the oblique. Thus, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} reflects more than abstract bias—it mechanistically represents asymmetry in stimulus sensitivity64. Observed orientation biases arise from both intrinsic bias (stimulus-independent growth rates) and stimulus sensitivity (how strongly the perceptual system responds to changes in stimulus properties). Vertical preference emerges because the neural population for vertical grouping is more sensitive to AR changes than the horizontal population. In the oblique axis, sensitivities for 45° and 135° percepts are nearly symmetrical, producing a weaker bias. Mechanistically, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} can be understood through a potential landscape metaphor65. Stable percepts correspond to valleys or attractors, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} determining the intrinsic tilt of the landscape. A higher γ value implies a more steeply tilted landscape for the cardinal axis, making the valley corresponding to vertical grouping deeper or more accessible by default. This represents more than a mere preference: it reflects a fundamental property of the system’s energy landscape that renders one state more probable than another. Such an effect could arise, for example, from a larger population of neurons tuned to vertical orientations or from stronger synaptic connectivity within that population. The mutual inhibition parameter (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}) was greater than 1 in both axes and higher in the cardinal axis. Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} quantifies the ratio of cross-inhibition to self-inhibition, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g>1$$\end{document} indicates stronger cross-inhibition, producing winner-take-all dynamics, bistability, and hysteresis. A higher \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} in the cardinal axis implies stronger competition between 0° and 90° percepts compared to 45° and 135°. In landscape terms, this corresponds to deeper wells separated by higher barriers, making switching harder and percepts more stable in the cardinal axis. Conversely, lower \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} in the oblique axis yields shallower wells and lower barriers. This suggests greater stability and hysteresis in the cardinal axis. However, behavioral data did not show differences in hysteresis between axes. At first glance, statistically equal hysteresis despite differing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} values seems paradoxical. The LVH model resolves this by showing that hysteresis magnitude depends not on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} alone but on the interaction between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}64,65. In the cardinal axis, strong competition (high \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document}) is counterbalanced by asymmetric sensitivities (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}), while in the oblique axis weaker competition is paired with near-symmetric sensitivities. These distinct dynamics converge on similar hysteresis magnitudes, revealing that equal behavioral outcomes can arise from fundamentally different neural mechanisms. Overall, the LVH model provides a plausible mechanistic account of perceptual switching, framing it as bifurcations in a dynamic system of competing neural populations. By estimating \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}, it uncovers latent dynamics not visible in behavior alone and provides explanatory power about neural mechanisms. Moreover, the model is predictive: For example, it can simulate how noise, different stimulus ranges, or learning might alter dynamics over time—capacities that descriptive fits of PSE or hysteresis cannot offer. In sum, our findings show that the LVH model not only accounts for orientation biases and perceptual switching in dot lattices but also reveals the hidden system parameters that generate them. This demonstrates its potential as a general mechanistic framework for understanding perceptual switching. This study had several limitations that should be considered when generalizing the results. To minimize the experimental duration, the number of unique stimuli was limited, though pilot results indicated a stable pattern of results. The study focused solely on transitions between the {0°, 90°} and {45°, 135°} angle pairs. Future research could examine switching between other pairs to better understand orientation effects in perceptual grouping under temporal influence. Additionally, this research used rectangular lattices within a circular aperture to investigate bistability. Future work could explore a broader range of dot lattices9, including those with more than two dominant stable organizations. Finally, while this study induced bistability by varying interdot distances at a fixed angle, future research could vary the stimulus angle, or both the angle and interdot distances simultaneously. This study examined temporal context in proximity-based grouping. Future research could extend this to other grouping principles, such as similarity and continuity, in dot lattices. Temporal effects on perceptual grouping could also be tested with other stimuli and paradigms, including Gabor lattices66, the central column grouping task67, and binocular rivalry68. The Lotka-Volterra-Haken amplitude equations were used here to model the data. Other computational approaches, such as attractor and oscillatory models69,70, could be applied to analyze multistability in dot lattices. This study employed a deterministic approach, whereas future work could use stochastic models that account for noise and adaptation71–73, particularly to capture trial-by-trial variability in responses. Supplementary Information Supplementary Information. Supplementary Information.