WorksPhysics and Astronomy
A Two-Population von Mises–Fisher Distribution of the Magnetic Field Direction in the Young Solar Wind
Abstract
The one-point distribution of the magnetic field direction in the young solar wind is found to be a superposition of two populations rather than a single exponential. Using Parker Solar Probe (PSP) Encounter 1 magnetometer data, we show that the distribution of the pitch-angle cosine mu(t) = bhat(t) . bhat_0 is well described by a two-component von Mises-Fisher (vMF) mixture: a broad, weakly confined component of concentration kappa_1 ~ 3.6 that accounts for the large-amplitude rotations and switchback tail, and a field-aligned component that reduces, in the small-angle limit, to a Gaussian in the polar angle theta of width sigma ~ 15 degrees (concentration kappa_2 ~ 14). The concentration of the field-aligned component is not a free parameter: it is set by the level of the transverse Alfvenic fluctuation through kappa_2 = 2<|B|^2>/<delta B_perp^2> = 2/<1-mu^2> ~ 16.2, in agreement with the fitted value to within ~12%. The transverse fluctuations are weakly anisotropic (<delta B_1^2>/<delta B_2^2> ~ 1.1), so the core is a weakly anisotropic Bingham distribution rather than a pure vMF. The broad component is exponential in mu, not in theta, confirming that switchbacks saturate toward mu = -1 rather than decaying away, consistent with the drift-diffusion framework of Bian et al. (2024, ApJ, 960, L15). This physics-based two-population framework connects macroscopic fluid parameters directly to those describing the directional statistics, offering a powerful diagnostic for probing the structural evolution of MHD turbulence and switchbacks in the inner heliosphere.
Email: [email protected]
Introduction
The magnetic field vector measured by a spacecraft in the turbulent solar wind deflects persistently from the mean field direction [22, 18]. Such fluctuations have been examined in the context of extension of coronal field into the heliosphere by Ness & Wilcox [20] (1967). At 1 au, the distribution of the angle between the instantaneous and the average field, over different solar wind conditions, have been examined in, e.g., [10, 11] (Xu et al. 2015). The entire spectrum of variations from the Parker field can be attributed to dynamical processes occurring in the plasma. In Alfvénic streams, the field magnitude remains close to a constant, so that the evolution of can be understood from that of the tip of the unit vector , which is confined to the unit sphere. The one-point statistics of is conveniently described by the pitch-angle cosine,
where is the polar angle measured from the mean field direction. The random walk of on the unit sphere was first treated by Barnes [5] as a pure rotational diffusion, which relaxes the field direction toward an isotropic distribution with no preferred direction. Bian & Li [8] generalized this to a steady-state advection–diffusion (drift–diffusion) model, in which a mean-field restoring force advects back toward against the stochastic diffusion; the balance of the two yields the exponential (equivalently, von Mises–Fisher, vMF) distribution
with concentration parameter related to the mean pitch-angle cosine by the Langevin function . The exponential form was established observationally by Bavassano & Mariani [6], Borovsky [9], and Dudok de Wit et al. [12]; for large the exponential in reduces to a Gaussian in , the standard fitting practice to the angular data at small angles.
A single exponential, however, provides only a first description. Two distinct features of the young solar wind suggest that a one-parameter fit is inadequate. First, the magnetic field direction exhibits a broad, nearly isotropic component associated with large-amplitude rotations and reversals of the radial field (switchbacks; [3, 13] (Fargette et al. 2021)), which populate the tail . Second, the bulk of the fluctuations are small-angle, nearly transverse deflections about the mean field ([7]), which produce a concentrated excess near . A natural and minimal description is therefore a two-population directional distribution. Two-population fits to the angular deflection of the magnetic field has been practiced by Xu et al. 2015, who fitted the angular distribution by the sum of a Gaussian core and an exponential tail in angle (instead of ).
In this Letter we carry out a two-population analysis of the PSP magnetic field data of Alfvénic streams within the directional-statistics framework of Bian & Li [8]. We show that has two vMF populations, one broad and one field-aligned. We also show that the concentration parameter of the field-aligned component vMF is regulated, without any free parameter, by the transverse Alfvénic fluctuation level. We clarify that the functional form of the broad component as exponential in , not in as in Xu et al. (2015).
Data and Method
We use the PSP/FIELDS fluxgate magnetometer Bale et al. [2], Level-2 RTN vectors at 1 minute resolution (PSP FLD L2 MAG RTN 1MIN), over Encounter 1 (2018 November 1–11), during which the heliocentric distance decreases from au to 0.17 au at perihelion and then returns to au. After discarding 22 NaNs, the series contains 15,818 data points. We separate the encounter into three intervals in heliocentric distance (Figure 1): the inbound leg au, the perihelion region au, and the outbound leg au. Within each interval the background field is a 6 hr moving average of the full vector field, and the pitch-angle cosine is with ; the mean pitch-angle cosine is .

Figure 1. PSP Encounter 1 (2018 November 1–11) overview. (a) Heliocentric distance (left axis) and heliographic latitude (right axis). (b) Proton plasma from SWEAP: velocity components (left axis) and number density (right axis, dotted). (c) The three RTN magnetic-field components and (d) the field magnitude , from PSP/FIELDS at 1 minute resolution. The shaded bands in (a) and (b) mark the three radial ranges: inbound au, perihelion au, and outbound au.
Figure 1 presents the macroscopic plasma and magnetic field conditions during Encounter 1. As the spacecraft descends from 0.25 au to a perihelion of au and back out, it remains immersed in a highly Alfvénic, relatively slow solar wind stream (– km s). The magnetic field is dominated by a sunward (negative) radial polarity. While the individual RTN magnetic field components (Panel c) exhibit intense, high-frequency fluctuations and frequent directional reversals (switchbacks), the total magnetic field magnitude (Panel d) evolves smoothly with radial distance and exhibits comparatively little high-frequency variance. This distinct contrast—violent directional scattering alongside a stable field magnitude—confirms that the fluctuations are predominantly transverse and spherically polarized.
We model the normalized distribution of by a mixture of two vMF (exponential) distributions,
where is the mass fraction of the broad component. For the second term reduces, using , to a Gaussian in the polar angle,
A total of 40 bins in from to , with a bin width of 0.05 were used in binning the data. We minimize the Baker-Cousins modified Poisson likelihood ratio between the model and the binned histogram of . Here is defined by Baker & Cousins [1],
where is the measured count in the -th angular bin, is the expected model count, and is the total number of equally spaced bins spanning .
To restrict the directional statistics to Alfvénic streams, we compute the normalized cross helicity and the residual energy from the SWEAP proton velocity and the Alfvén-normalized magnetic field ,
where and are the fluctuations about the local mean within a window. Because the proton velocity is reliable only at 1 hour resolution in the COHOWeb merged product, and are evaluated over 6 hr windows at 1 hr cadence (full RTN vector). A window is deemed to be Alfvénic if and we find that 212 of 258 windows (82%) are Alfvénic. The background field used to form is a 6 hr boxcar average of the full vector field, with the data point at the center of the window. A one-minute point is retained (“tallied”) only if more than half of its 6 hr window is Alfvénic. As a result, 12,914 of 15,430 points (84%) survive this selection criterion.
Figure 2 shows versus (colored by heliocentric distance) and their time series. The interval is strongly Alfvénic: the fluctuations lie near the circle, with mean and mean , i.e., near kinetic–magnetic equipartition with a weak magnetic excess. Binning the tallied points into the three radial ranges (Figure 1) gives 2995, 5332, and 2523 points (92%, 93%, and 77% of each range), and the two-component fit shows a weak radial trend as we discuss in the next section.

Figure 2. Alfvénicity of PSP Encounter 1 from 6 hr windows at 1 hr cadence (full RTN vector; ). (a) Residual energy versus normalized cross helicity , colored by heliocentric distance , with the unit circle (dashed) and the Alfvénicity threshold (dotted); most points cluster near the circle, confirming that this interval is strongly Alfvénic. (b) (blue) and (red) versus time.
Results and Discussion
Figure 3 shows the binned distribution of together with the two-component fit, in three ranges of heliocentric distance separating the inbound and outbound legs of the encounter, plus the combined sample. By separating the tallied data into the three heliocentric radial ranges, we observe a distinct spatial evolution in the two-population structure, reflecting the radial development of the turbulence (see Figure 3). Approaching the Sun, the core concentration drops significantly from the inbound leg () to the perihelion region (), indicating a broadening of the field-aligned core as the relative transverse fluctuation level intensifies deeper in the corona. Concurrently, the broad tail component also widens at perihelion () compared to the inbound leg (), corresponding to an increased occurrence rate and angular spread of switchbacks. The outbound leg exhibits intermediate concentrations (, ), which may reflect intrinsic longitudinal asymmetries or different solar wind source streams crossed by the spacecraft during its trajectory. Despite this radial variation, the two-component vMF mixture accurately captures the distribution across all distances, confirming its universality in the young solar wind.

Figure 3. Two-component fit to the pitch-angle cosine (log–linear) for the Alfvénic periods of PSP Encounter 1, in three ranges of heliocentric distance (inbound au, perihelion au, outbound au) and, in the bottom-right panel, all three combined. In each panel the broad exponential tail (red dashed line) dominates the switchback tail and the Gaussian-in-angle core (blue dashed line) dominates near ; the fitted , , and the number of points are given in each panel.
On the combined data, the broad component has concentration and fraction and the field-aligned core component has (equivalent to a width of ) and a fraction of . The two-component fit gives over the full range . The broader component of this mixture () include in it the ubiquitous magnetic field switchbacks observed extensively in the inner heliosphere. These structures represent large-amplitude, nonlinear Alfvénic fluctuations evolving on top of the background field. Large-amplitude spherically polarized Alfvén waves have been established in earlier foundational work by Barnes [4]. As the exact nonlinear solutions to the ideal magnetohydrodynamic equations, these are fluctuations possessing a strictly conserved magnetic field magnitude (). Later, Barnes [5] developed a statistical theory for Alfvénic disturbances, in which the magnetic field moves randomly over the surface of a sphere of radius . When incorporating these insights into modern interpretations of Parker Solar Probe observations, the broader tail component () of the angular distribution naturally captures the large-amplitude arc-polarized nature of these waves. Recently Bian & Li [8] showed that these structures occur naturally as the steady state solution of an advection-diffusion process of the tip of the B field on a sphere, extending the work by Barnes [5].
Comparing to the broad component, the core component is fully governed by the small-amplitude transverse Alfvénic fluctuations, and can be related to small-amplitude transverse Alfvénic fluctuation level. This can be seen from the following. Let the unit vectors span the plane perpendicular to the mean direction , with in the radial–tangential (RT) plane and along the normal. The transverse deflection of the field is then,
with . Note that is the cosine of the angle between and , and spherical polarization still holds. If we model the small-amplitude Alfvénic fluctuation as two independent Gaussian random variables, along the two transverse directions and with root-mean-square ( are dimensionless), then we have
Note that the tangent-plane area element coincides with the solid-angle element at small angles (). Eq. (8) is the Bingham distribution on the unit sphere when .
Assuming , the Bingham distribution collapses to the vMF distribution,
where at small angles. Because (equal to in the isotropic case), therefore the core component is governed by the transverse fluctuation level,
In the last step, we make use of the fact that the field magnitude is nearly constant for Alfvénic fluctuations, and compute it using .
As a consistency check, the posterior-weighted average of the core is 0.124 for the entire encounter, yielding . Alternatively, the posterior-weighted average and , giving a relative transverse fluctuation level , hence from the second part of equation (10). Both agree with the fitted to within , and this agreement holds in each radial range, the fitted vs that using are: (19.9 vs 21.8 for inbound, 10.7 vs 12.5 for perihelion, 16.2 vs 17.9 for outbound; see Figure 3). We remark that the posterior weights used in obtaining are drawn from the two-component fit, therefore the above calculation is not fully independent of the fit. Nevertheless, the agreement of between equation (10) and the fitting provides a meaningful consistency check, as it indicates that the core concentration follows from the transverse-fluctuation moments. In part, the discrepancy between the estimation and the fitting of is due to the fact that there is a weak transverse anisotropy between and , so that the exact core is a Bingham rather than a vMF.
The emergence of a vMF distribution is explained using the Fokker-Planck framework by Bian & Li [8], where a drift diffusion process with a constant restoring rate is assumed, yielding the one-population exponential of Eq. (2). The observed two-population structure indicates that a constant restoring rate is insufficient. A natural extension of Bian & Li [8] is to consider a -dependent restoring rate which is larger near (strongly focusing the small-angle core) and decreases as decreases, accommodating the observed switch-back populations.
The geometric and physical basis for this state-dependent restoring force is visualized in Figure 4, which illustrates a plasma parcel embedded in a spherically polarized Alfvénic wave packet. Because the plasma is tied to the physical field line, its transverse velocity fluctuation is parallel to the magnetic perturbation and strictly perpendicular to the total local field . In the constant- spherical phase space, this transverse motion naturally decomposes into azimuthal () and polar () directions. The azimuthal component represents phase-mixing and does not contribute to the restoring torque. The polar restoring torque , however, drives the relaxation of the pitch angle and is composed of two distinct spatial gradients. As demonstrated below, the linear background tension dominates at small deflection angles (acting over the parallel correlation length ), while the nonlinear self-tension becomes dynamically significant at large deflection angles (acting over a length scale along , which we denote as ). This physical decomposition reveals why the macroscopic relaxation rate is fundamentally -dependent, providing the origin for the two distinct populations in the observed angular distribution.

Figure 4. 3D geometric visualization of a plasma parcel (black dot) embedded in a spherically polarized Alfvénic wave packet. The physical field line (green) restricts the local magnetic field (blue) and its perturbation (orange) to a constant- sphere. The macroscopic restoring torque (red) drives the relaxation of the pitch angle . This torque is composed of two competing spatial gradients: the linear background tension acting over the parallel correlation length (purple), and the non-linear self-tension acting over the active perturbation scale (magenta). As detailed in the text, the dynamic competition between these two length scales determines the state-dependent relaxation rate .
We now show that a -dependent restoring force can be understood by considering a fluid parcel’s motion in Alfvénic fluctuations. Let the parcel have a mass . The parcel’s motion is governed by Newton’s second law, driven by three primary forces: a macroscopic restoring force from magnetic tension , a macroscopic turbulent drag or damping from the surrounding plasma, (see (Li & Bian 2023)), and a rapidly fluctuating stochastic force representing the random impacts of turbulent eddies ():
For Alfvénic fluctuations, the parcel velocity and the magnetic perturbation are connected by the Walén relation: , and due to the constant- spherical polarization constraint, the field vector is confined to the surface of a sphere. We can therefore uniquely parameterize both the displacement of the plasma parcel and the deflection of the magnetic field vector by the polar angle relative to the mean field . The transverse velocity (with respect to the total field) is directly proportional to the angular drift (). Furthermore if we assume the drag from turbulent scattering instantly overwhelms the parcel’s inertia, then the acceleration term vanishes and we have,
Because the transverse velocity fluctuation and the magnetic perturbation are parallel, the angular coordinate tracking the fluid parcel’s position is geometrically identical to the pitch angle between the total field and the mean field . The generalized angular force driving the Langevin equation is simply the polar projection of the macroscopic magnetic tension force density, , onto the tangent of the constant- sphere. Decomposing the field into a background and a fluctuation (), the tension force splits into a linear and a non-linear component:
To evaluate these spatial gradients, we follow standard phenomenological scaling for MHD turbulence ([15]), and assume the two terms operate over fundamentally different correlation length scales (Figure 4).
The linear term represents the spatial derivative strictly along the mean field . Approximating this gradient over the parallel correlation length , the background tension acts as a restoring force pulling the perturbation straight, yielding a 3D force . Projecting this physical pull onto the spherical tangent vector isolates the effective macroscopic restoring torque, which is strictly proportional to the sine of the deflection angle:
Conversely, the non-linear self-tension term represents the gradient along the fluctuation itself. In turbulence phenomenology, variations along this active perturbation occur over its own correlation scale, (which replaces the strictly transverse eddy scale because is not perpendicular to at large amplitudes), giving the scalar operator . Because spherical polarization rigorously constrains the fluctuation magnitude to , applying this operator yields a non-linear force magnitude . When projected onto the meridian, this self-tension acts to inflate the folded field line, providing a counter-torque that opposes the linear background tension:
Modeling the turbulent impacts as Gaussian white noise requires care, as the fluctuations are constrained to the constant- spherical surface. For Brownian motion on a Riemannian manifold with metric , the Itô stochastic differential equation requires a Laplace-Beltrami geometric drift [17, 16]:
where is the local coordinate of the curved space or manifold (for a sphere, these are the polar angle and azimuthal angle ), is the inverse metric tensor, is the determinant of the metric tensor, and is the dispersion matrix, strictly defined by the manifold’s geometry through the relation . In our case, the metric determinant generates a purely geometric drift of , reflecting the larger phase-space volume available toward the equator. Substituting this along with the combined torque into Equation (12) yields the complete Langevin equation:
where is the Wiener increment and the state-dependent macroscopic relaxation rate emerges explicitly from the competition between the two length scales:
To directly compare this physical derivation with the statistical framework of Bian & Li [8], we map the angular dynamics into the pitch-angle cosine . Applying Itô’s Lemma, , the geometric drift and the Itô correction exactly combine to yield a noise-induced drift of . The Langevin equation in -space is therefore:
Comparing this to the generic drift-diffusion model utilized by Bian & Li [8], we identify the diffusion coefficient and the drift coefficient . Setting the angular diffusivity , the concentration parameter , and using the dimensionless time , equation (19) maps directly to Equation (13) of [8].
While Bian & Li [8] assumed a constant to obtain a single vMF distribution, our derivation here shows that is -dependent. The steady-state Fokker-Planck solution is . Because the physical transition between the linear-dominated and non-linear-balanced regimes is stark, it effectively bifurcates the phase space. Rather than performing a complex continuous integration, approximating this solution as a superposition of two constant- asymptotic states naturally recovers the observed two-component vMF mixture.
In the small-angle limit (), the non-linear term vanishes and the magnetic tension is dominated by the linear background field operating over . This rigid restoring force provides a rapid relaxation rate, , driving the field direction into the highly concentrated vMF core population (). As the wave amplitude grows to large angles (), the non-linear self-tension becomes dynamically significant. Rather than overwhelming the background field and produce an unphysical repulsive force, the spatial scales reorganize (e.g., shrinks to become comparable to ) such that the internal self-tension counter-balances the linear background tension, allowing the structure to approach local force-balance. This cancellation weakens the net restoring torque, collapsing the macroscopic relaxation rate to a much smaller residual value, . Consequently the field direction lingers in these large-angle and reversed states, sustaining switchbacks and yielding the broad, weakly confined tail population ().
Summary
In summary, we have shown that the one-point angular distribution of the magnetic field in the highly Alfvénic young solar wind is fundamentally a two-population von Mises-Fisher (vMF) mixture, rather than a single exponential. Using PSP Encounter 1 data, we demonstrated that the distribution consists of a highly concentrated, field-aligned core and a broad, weakly confined tail. The core component is rigorously constrained by the transverse Alfvénic fluctuation level, yielding a concentration parameter that matches observations without any free fitting parameters. Meanwhile, the broad tail component accurately captures the large-amplitude switchback population, confirming that these structures saturate toward in a steady-state drift-diffusion balance.
The physical origin of these two distinct populations lies in the state-dependent screening of the macroscopic magnetic tension force. By decomposing the restoring torque into a linear background term and a non-linear self-tension term, we revealed a dynamic competition governed by two distinct correlation length scales. At small deflection angles, the rigid linear tension operating over the parallel scale provides a strong, constant restoring rate, dynamically confining the core population. At large angles, however, the structure approaches local force-balance as the non-linear self-tension, acting over the active perturbation scale , grows to directly counter-balance the background tension. This internal cancellation effectively collapses the macroscopic relaxation rate, allowing the field direction to linger in reversed states and naturally sustaining the long-lived switchback tail. This physics-based two-population framework connects macroscopic fluid parameters ( and ) directly to those describing the directional statistics ( and ), offering a powerful diagnostic for probing the structural evolution of MHD turbulence and switchbacks in the inner heliosphere.
Acknowledgments.—This work is supported, at MUST, in part by FDCT grants 0022/2025/ASJ, 0137/2025/AFJ, and FRG-25-113-SSI, and by the São Paulo Research Foundation (FAPESP), Process Number 2025/11756-6. RAM acknowledges support from CNPq (407341/2022-6, 407493/2022-0), FAPDF (383/2023) and Brazilian Space Agency (TED 000529/2024). ELR acknowledges support from CNPq (306265/2024-9). B.R. was supported by FAPESP under grants 2023/16513-9 and 2024/20546-2.
References
- [1]Baker, S., & Cousins, R. D. 1984, Nuclear Instruments and Methods in Physics Research, 221, 437, doi: 10.1016/0167-5087(84)90016-4
- [2]Bale, S. D., Goetz, K., Harvey, P. R., et al. 2016, Space Science Reviews, 204, 49, doi: 10.1007/s11214-016-0244-5DOI
- [3]Bale, S. D., Badman, S. T., Bonnell, J. W., et al. 2019, Nature, 576, 237, doi: 10.1038/s41586-019-1818-7DOI
- [4]Barnes, A. 1976, Journal of Geophysical Research, 81, 281, doi: 10.1029/JA081i001p00281DOI
- [5]Barnes, A. 1981, Journal of Geophysical Research, 86, 7498, doi: 10.1029/JA086iA09p07498DOI
- [6]Bavassano, B., & Mariani, F. 1983, in NASA Conference Publication, Vol. 228, NASA Conference Publication, ed. M. Neugebauer, 0..99
- [7]Belcher, J. W., & Davis, Jr., L. 1971, Journal of Geophysical Research, 76, 3534, doi: 10.1029/JA076i016p03534DOI
- [8]Bian, N. H., & Li, G. 2024, The Astrophysical Journal Letters, 960, L15, doi: 10.3847/2041-8213/ad17ccDOI
- [9]Borovsky, J. E. 2008, Journal of Geophysical Research: Space Physics, 13, A08110, doi: 10.1029/2007JA012684DOI
- [10]Borovsky, J. E. 2010, Journal of Geophysical Research (Space Physics), 115, A09101, doi: 10.1029/2009JA015040DOI
- [11]Chang, Q., Xu, X., Wang, X., et al. 2022, The Astrophysical Journal, 931, 105, doi: 10.3847/1538-4357/ac6bf3DOI
- [12]Dudok de Wit, T., Krasnoselskikh, V. V., Bale, S. D., et al. 2020, The Astrophysical Journal Supplement Series, 246, 39, doi: 10.3847/1538-4365/ab5853DOI
- [13]Fargette, N., Lavraud, B., Rouillard, A. P., et al. 2022, Astronomy and Astrophysics, 663, A109, doi: 10.1051/0004-6361/202243537DOI
- [15]Goldreich, P., & Sridhar, S. 1995, The Astrophysical Journal, 438, 763, doi: 10.1086/175121DOI
- [16]Hsu, E. P. 2002, Graduate Studies in Mathematics, Vol. 38, Stochastic Analysis on Manifolds (Providence, RI: American Mathematical Society)
- [17]Ikeda, N., & Watanabe, S. 1981, North-Holland Mathematical Library, Vol. 24, Stochastic Differential Equations and Diffusion Processes (Amsterdam: North-Holland)
- [18]Jokipii, J. R. 1966, The Astrophysical Journal, 146, 480, doi: 10.1086/148912DOI
- [20]Ness, N. F., & Wilcox, J. M. 1966, The Astrophysical Journal, 143, 23, doi: 10.1086/148473DOI
- [22]Parker, E. N. 1958, The Astrophysical Journal, 128, 664, doi: 10.1086/146579DOI