Partial coherence control delivers skyrmionic topological resilience and transitions Authors: Yonglei Liu, Shiqi Chen, Zhenyu Guo, Kaiqi Zhu, Yahong Chen, Yangjian Cai, Yijie Shen, Fei Wang Published: 2026 ABSTRACT AbstractOptical skyrmions have recently unlocked topological quasiparticle textures of light, rising in prominence for next-generation ultra-robust information processing. However, to date, their study has been mainly confined to coherent laser fields. Here we extend skyrmions to more general light sources of partially coherent, stochastic optical fields. We define stochastic optical skyrmions and uncover a hidden regime where spatial coherence acts as a primary determinant of topological stability. While environmental randomness typically degrades fully coherent states, we demonstrate that engineered partial coherence provides a self-healing mechanism that preserves topology under extreme turbulence. Moreover, we show that the coherence structure can be actively tailored to trigger on-demand topological phase transitions, such as skyrmion-to-skyrmionium conversion and skyrmion lattice splitting. These findings redefine the boundaries of topological photonics, paving the way for resilient and high-fidelity information platforms that remain operational in general, non-ideal, real-world environments.Subject terms: Optical physics, Applied optics, Optical physics Introduction Skyrmions are topologically protected quasiparticles first introduced in nuclear physics1,2. They have become an important concept in condensed matter physics3,4, especially in magnetism5,6, where their robustness and reconfigurability support emerging information technologies7,8. Recently, the skyrmion paradigm has been extended to electromagnetic waves9–11, including free-space optical beams12–15, enabling photonic realizations of topological textures with growing relevance to structured-light control16–22, optical computing23, and high-dimensional information processing24,25. Analogous to magnetic skyrmions, optical skyrmions have been shown to exhibit a degree of topological protection against external perturbations in harsh environments, including complex media, biological tissues, and turbulent atmospheres26–31. However, under strong atmospheric turbulence, the skyrmion number, the key topological invariant used to quantify the skyrmionic texture, can degrade and become realization-dependent. Recovering the correct skyrmion number may then require long integration times29 or extensive post-processing30–32, which is incompatible with scenarios that demand rapid, real-time readout. This limitation is particularly restrictive for practical applications, such as topology-encoded free-space optical communications, where information is carried by the skyrmion number and must be decoded on the fly. A natural route to overcome this challenge is to move beyond idealized laser fields33. Nearly all optical skyrmions reported to date have been created and studied using fully coherent, fully polarized beams. Yet natural optical fields are intrinsically stochastic: partial coherence and random perturbations are the rule rather than the exception34,35. Partially coherent beams have been widely studied and have shown distinct advantages in suppressing turbulence-induced fluctuations, improving beam transport, and enhancing the robustness of free-space optical communications33,36. This raises a fundamental question: are optical skyrmions a generic feature of partially coherent light, and can partial coherence endow them with enhanced resilience to environmental randomness, such as atmospheric turbulence? Here we establish the concept of partially coherent optical skyrmions. We demonstrate stochastic skyrmion textures and show that their topology can be preserved during propagation through spatial-coherence engineering. Remarkably, partially coherent skyrmions remain robust even under strong turbulence—without requiring long integration times or extensive post-processing—marking a key step toward real-time, topology-encoded photonic links in complex environments. Moreover, we uncover coherence-driven topological transitions in the low-coherence regime, including skyrmion-to-skyrmionium conversion and the controlled evolution from isolated skyrmions to ordered skyrmion lattices. Together, these results extend optical skyrmions from ideal coherent laser fields to the more universal domain of partially coherent light, opening routes to turbulence-resilient topological photonics for free-space communications, sensing, and photonic information processing. Results Concept Skyrmion textures in optics arise from a continuous, topological mapping between an internal parameter space and the real-space distribution of an optical field11, as schematically illustrated in Fig. 1a. Such textures have been constructed using various degrees of freedom of light, including the electric field9, spin angular momentum10, Poynting vector18, and even scalar evanescent components21. For fully coherent beams, the most widely explored realization is the Stokes skyrmion12, in which the parameter space is the Poincaré sphere spanned by the (normalized) Stokes vector S(r), and the corresponding real-space texture is encoded in the spatially varying polarization state across the transverse beam profile. A canonical platform is provided by full Poincaré beams37, whose transverse field can be constructed as a coherent superposition of two orthogonally polarized Laguerre–Gaussian (LG) modes with, in general, different radial and azimuthal indices, where r is the two-dimensional coordinate in a transverse plane, perpendicular to the beam propagation axis. LGpl denotes the complex amplitude of an LG mode with radial index p and azimuthal index l, a1,2 ≥ 0 set the relative modal weight (with amplitude ratio η = a2/a1), γ is the relative phase, and e^1,2 are two orthogonal unit polarization basis vectors. When ∣l1∣ ≠ ∣l2∣, the field in Eq. (1) supports skyrmion features in the Stokes texture S(r) (see Methods). The topology of a given skyrmion configuration is quantified by the skyrmion number11 where σ denotes the region over which the skyrmion is defined. Physically, Nsk counts how many times the Stokes vector wraps the unit parameter sphere as r spans σ. Concept and generation of partially coherent optical skyrmions. a Optical skyrmions correspond to a topological mapping between the real-space polarization texture and a parameter space represented by the Poincaré sphere. b A partially coherent skyrmion beam is generated by transmitting a fully coherent (deterministic) skyrmion through a temporally varying, polarization-insensitive random phase screen T(r) and performing an ensemble (time) average. c At the source plane (z = 0), the field is fully polarized with P(r) = 1, so the Stokes vectors lie on the unit Poincaré sphere. d After propagation (z > 0), spatial partial coherence leads to inhomogeneous depolarization (0 ≤ P(r) ≤ 1), causing the Stokes-vector locus to shrink into the interior of the unit sphere. e A reshaped parameter-space representation is obtained by local renormalization of the Stokes vector, S¯(r)=S(r)/P(r), which restores a unit-sphere mapping for characterizing the skyrmion topology. To generate a partially coherent skyrmion field, as illustrated in Fig. 1b, one may conceptually pass a fully coherent skyrmion field through a temporally varying, polarization-insensitive random phase screen. Such a dynamic screen imprints a time-dependent, spatially random phase on the beam, thereby reducing the spatial coherence while preserving the underlying polarization structure38,39. This picture is closely related to free-space propagation through atmospheric turbulence40, which is often modeled as a sequence of polarization-insensitive random phase screens. For statistically stationary screens, the second-order statistical properties of the resulting partially coherent vector field are described by the cross-spectral density (CSD) matrix34 where μ(r1, r2) is the complex degree of spatial coherence between the points r1 and r2, and * and T denote complex conjugation and transpose, respectively. The spatial structure of μ(r1, r2) is determined by the correlation of the random screen, i.e., μ(r1, r2) ∝ 〈T*(r1)T(r2)〉, where 〈 ⋅ 〉 denotes a time average and T(r) is the screen transmission function. In the experiment, however, the partially coherent skyrmion fields are not produced by first creating a deterministic coherent skyrmion and then passively decohering it in a fluctuating medium. Instead, they are synthesized directly at the source plane using the random-mode superposition method. In this approach, each random realization contains the prescribed vector-mode structure, while the ensemble of realizations determines the designed spatial coherence function. Thus, both the skyrmionic topology and the coherence properties are engineered at the source. Experimental details are provided in Methods. Equation (3) implies that the partially coherent skyrmion source shares the same ensemble-averaged polarization structure as its fully coherent counterpart in Eq. (1). In particular, because the random phase screen is polarization-insensitive, it does not alter the averaged polarization matrix at the source plane, and hence yields the same Stokes texture S(r). As a result, both the fully coherent and the partially coherent fields exhibit the same skyrmion configuration at the source: the Stokes vectors lie on the surface of the unit Poincaré sphere, as illustrated in Fig. 1c. Upon propagation, however, the presence of spatial partial coherence at the source leads to a reduction of the local degree of polarization at the observation plane38,39. Consequently, the measured Stokes vectors no longer remain on the unit sphere but shrink into its interior, reflecting a partially polarized field. Importantly, unlike the case of a spatially uniform (global) mixed state27, the depolarization induced by spatial partial coherence is generally inhomogeneous: the degree of polarization P(r) varies across the beam profile. Geometrically, this corresponds to a nonuniform inward “denting” of the Stokes-vector locus on the Poincaré sphere, with the radial distance to the sphere center given by P(r), as sketched in Fig. 1d. To define an appropriate parameter-space representation for such inhomogeneously partially polarized skyrmion fields, we locally renormalize the Stokes vector as so that ∣S¯(r)∣=1 pointwise. This normalization restores a unit-sphere mapping and provides a well-defined parameter sphere for characterizing the skyrmion texture, as shown in Fig. 1e. Accordingly, the skyrmion number is evaluated from Eq. (2) by replacing S(r) with S¯(r). It is important to note that Eq. (4) becomes ill-defined when the local degree of polarization approaches zero, i.e., P(r) → 0. Physically, this limit corresponds to a locally unpolarized field, for which the polarization state is completely randomized and no well-defined local Stokes vector can be assigned. In other words, as P(r) decreases, the skyrmionic Stokes texture encoded in the partially coherent field is progressively buried in the unpolarized background; in the limiting case P(r) = 0, the local skyrmionic texture disappears. In extreme situations, such as very low spatial coherence or very strong turbulence, the degree of polarization may be strongly degraded. In this case, the topological texture may be overwhelmed by the unpolarized background and random noise, and the skyrmion number can no longer be reliably extracted without additional signal recovery. Experimentally, such weak polarization textures could in principle be accessed by improving the polarization signal-to-noise ratio, for example, through higher-sensitivity polarimetric detection, longer averaging over source fluctuations, or other polarization-signal recovery techniques. These extreme cases, however, lie outside the parameter regime investigated in the present work. We further note that the definition in Eq. (4) is generally applicable to partially coherent and partially polarized optical fields, irrespective of their specific spatial coherence structure, provided that the local degree of polarization remains finite in the region where the topology is evaluated. Partially coherent skyrmion beam To investigate the propagation dynamics of partially coherent skyrmion beams, we first synthesize a skyrmion field with a Gaussian degree of spatial coherence35, i.e., μ(r1,r2)=exp−(r1−r2)2/(2δ02), where δ0 controls the transverse coherence width. Experimental details of the synthesis and characterization are provided in Methods. The input field is formed by superposing a right-handed LG01 mode and a left-handed LG00 mode in two orthogonal polarization channels, with equal modal weights η = 1, see in Fig. 2a, and with the coherence width of δ0 = 0.5 mm. Figure 2b–e shows the measured Stokes parameters and the corresponding Stokes texture S¯(r) of a partially coherent skyrmion beam during focusing by a thin lens. The corresponding theoretical results are given in Supplementary Sec. S3. The focal plane represents the far-field distribution of the beam. At the source plane, the Stokes texture exhibits a Néel-type skyrmion, with a measured skyrmion number Nsk = − 0.8789. The deviation from the ideal value Nsk = − 1 arises from the finite support used in the experimental evaluation: Nsk is computed over the region where the intensity exceeds 2% of the peak value, whereas the theoretical value corresponds to integration over the entire transverse plane. Propagation dynamics of a partially coherent skyrmion beam with a Gaussian degree of spatial coherence. a The input polarization texture is synthesized by coherently superposing two orthogonally polarized modes: a right-handed LG01 and a left-handed LG00. b–e Measured transverse distributions of the unnormalized Stokes parameters s0–s3 (top rows) and the corresponding normalized Stokes texture S¯(r) (bottom rows) at the source plane and at selected propagation distances z during focusing by a thin lens with focal length f. The skyrmion number Nsk extracted from each texture is indicated below. Evolution of (f) the skyrmion number Nsk and (g) the polarized power fraction Pp, defined as the ratio of the fully polarized power to the total power, as a function of propagation distance for different coherence widths δ0. Upon propagation from the source plane to the focal plane, the transverse patterns of the (unnormalized) Stokes parameters s1 and s2 undergo a clockwise rotation by approximately π/2, which transforms the Stokes texture from the Néel-type into a Bloch-type configuration. Importantly, the skyrmion number remains effectively unchanged throughout this evolution, consistent with its topological character. By varying the spatial coherence width of the beam, this Néel-to-Bloch conversion is observed to follow the same evolution. Although reducing the spatial coherence leads to progressive depolarization during propagation, as evidenced in Fig. 2g by the decrease of the polarized power fraction Pp from unity, the skyrmion number Nsk, defined via the renormalized Stokes field, remains effectively stable, as shown in Fig. 2f. The slight decrease in Nsk observed during propagation is mainly attributed to the finite-intensity region used in the experimental evaluation. We find that the free-space robustness of partially coherent skyrmion textures depends sensitively on the amplitude ratio η between the two orthogonally polarized constituents. For example, at η = 1.25 [Fig. 3a–d], the skyrmion number drops from Nsk = −0.9226 at the source plane to Nsk = 0.1296 at the focal plane, indicating an abrupt loss of the skyrmion texture. This strong η dependence can be understood from the distinct propagation behaviors of the partially coherent LG01 and LG00 modes in the two polarization channels. Unlike a fully coherent LG01 beam, which maintains a dark on-axis core, a partially coherent LG01 beam gradually builds up on-axis intensity during propagation41. As a result, beyond a critical distance, the on-axis intensity of the partially coherent LG01 component can exceed that of the partially coherent LG00 component [see Fig. 3e]. Once this crossover occurs, the sign of the central S¯3 reverses from − 1 to 1 [see Fig. 3f], triggering an abrupt breakdown of the skyrmion topology. Consistently, Fig. 3g shows that S¯3(r) no longer spans the full range [ − 1, 1] at the corresponding propagation distance, so the Stokes mapping fails to cover the entire parameter sphere and the topological texture becomes corrupted. Notably, the above physical mechanism also applies to partially coherent skyrmion beams with higher-order azimuthal indices, i.e., when both l1 and l2 are nonzero. A detailed analysis of the critical propagation distance is provided in Supplementary Sec. S1. Amplitude-ratio-dependent robustness and breakdown of partially coherent skyrmion textures. a–d Evolution of the normalized Stokes texture S¯(r) at the source plane and at selected propagation distances during focusing (Gaussian spatial coherence), shown here for η = 1.25. The indicated labels “down” and “up” mark the sign of the central S¯3 component, and the skyrmion number Nsk extracted from each texture is given below. e Evolution of the on-axis intensities of the partially coherent LG00 and LG01 components during propagation for three different amplitude ratios. f Evolution of the central Stokes component S¯3(0,0) during propagation. The critical propagation distances at which the on-axis intensity of the partially coherent LG01 component exceeds that of the LG00 component, leading to the reversal of S¯3(0,0), are highlighted by circles. g Cross-sectional profiles of S¯3(x,0) at z = 0.95f for representative amplitude ratios η = 1.00, 1.25, and 2.50, illustrating the on-axis crossover that leads to a central S¯3 sign reversal when η > 1 (insets: corresponding S¯3(r) maps). h Measured skyrmion number Nsk as a function of propagation distance and amplitude ratio η, showing accelerated degradation of topology for η > 1 and robust preservation for η ≤ 1. As η increases further, the decay of Nsk during focusing becomes progressively faster, as summarized in Fig. 3h. By contrast, for η ≤ 1, the skyrmion texture remains intact throughout propagation because the on-axis intensity of the LG00 component stays larger than that of the LG01 component [Fig. 3e], preventing the central S¯3 inversion [Fig. 3f]. Additional theoretical results for different η are given in Supplementary Fig. 1. Further experimental and theoretical propagation dynamics for other partially coherent skyrmion textures are presented in Supplementary Figs. 2–4, confirming the stability of partially coherent skyrmion beams in free space under the condition η ≤ 1. Robustness of partially coherent optical skyrmions in turbulence In recent years, increasing attention has been paid to the stability of skyrmion topological textures when structured beams propagate through harsh environments, such as atmospheric turbulence26,28–31. For a fully coherent skyrmion beam, one may naively expect the skyrmion number to remain invariant under turbulence, since turbulence is often modeled as a continuous phase perturbation. Here we show that this expectation can fail in the strong-turbulence regime, where diffraction accumulated along the turbulent channel cannot be neglected40. In this case, turbulence-induced diffraction can generate additional singular structures, leading to pronounced degradation and realization-to-realization instability of the skyrmion number. We note that, under strong turbulence, the skyrmion number can be partially recovered by ensemble averaging over many turbulence realizations29 or by extensive post-processing of the measured data30–32. However, such approaches are time-consuming and therefore impractical for applications that require rapid, real-time topology decoding in complex environments. In what follows, we demonstrate that partially coherent skyrmion beams can deliver a stable realization-to-realization skyrmion-number output even under strong atmospheric turbulence. The key point is that, in the partially coherent case, the detector records an ensemble-averaged Stokes texture over many source realizations within a single quasi-static turbulence realization. This corresponds to the time-scale condition τs ≪ τd ≪ τt, where τs is the fluctuation time of the partially coherent source, τd is the detector integration time, and τt is the characteristic evolution time of the turbulence. Under this condition, the detector averages over the fast source fluctuations but not over different turbulence realizations. Thus, a single set of Stokes measurements within a single turbulence realization is sufficient to determine a stable skyrmion number. This single-realization stability is the practical advantage of partial coherence in our work. This distinction highlights the central message of our work: whereas previous studies have mainly explored the robustness of optical topology to randomness29, here we show that engineered randomness, in the form of controlled partial coherence, can itself serve as a resource for stabilizing optical skyrmion topology. In this sense, our work addresses not only “robustness to randomness”, but also “robustness via randomness”. We first simulate a fully coherent skyrmion beam synthesized from two orthogonally polarized LG00 and LG02 modes propagating through a 1000 mm turbulent channel. The turbulence is modeled with a multi-phase-screen approach42,43 that explicitly includes diffraction between successive screens (see Supplementary Sec. S4). At the receiver plane we evaluate the Stokes parameters, the skyrmion number, and the complex Stokes field S¯c=S¯1+iS¯229 together with the corresponding mapping of S¯(r) on the reshaped parameter sphere. The phase singularities in the complex Stokes field, which occur at points where S¯1=S¯2=0 and hence S¯3=±1, correspond to circular-polarization singularities, i.e., C-points. These singularities are directly connected to the skyrmion number through Stokes’ theorem44. Although the phase modulation on each screen is continuous, Fig. 4a–c show that diffraction produces additional phase singularities in arg(S¯c) at the output plane as the turbulence strength increases (parameterized by the refractive-index structure constant Cn2). These extra singularities are accompanied by an increasingly irregular and fragmented mapping of S¯(r) on the parameter sphere, and correlate with a clear degradation of the skyrmion number. Under strong turbulence, e.g., Cn2=1×10−8m−2/3, different turbulence realizations yield substantial shot-to-shot fluctuations: as shown in Fig. 5a, the output skyrmion number becomes unstable with mean N¯sk=−1.6497 and standard deviation σ = 0.6431. We note that the parameter-sphere coverage plots and the phase-singularity maps of S¯c provide complementary views of the same topological degradation process. The former visualizes the global mapping from real space to the parameter sphere, whereas the latter reveals the underlying singular skeleton that governs the phase winding and contributes to the skyrmion number. Turbulence-induced singularity proliferation and coherence-enabled stabilization of skyrmion topology. Columns correspond to increasing turbulence strength (from left to right): without turbulence, Cn2=5×10−9m−2/3, and Cn2=1×10−8m−2/3. Rows correspond to decreasing spatial coherence (from top to bottom): fully coherent input, partially coherent input with δ0 = 1.0 mm, and δ0 = 0.5 mm. For each case, each panel shows the phase of the complex Stokes field arg(S¯c) with S¯c=S¯1+iS¯2 (left) and the corresponding mapping of the renormalized Stokes vectors S¯(r) on the reshaped parameter sphere (right). The extracted skyrmion number Nsk is indicated above each panel. With increasing turbulence, diffraction generates additional phase singularities in S¯c and leads to an increasingly irregular parameter-sphere mapping for the fully coherent beam (a–c). Reducing the spatial coherence suppresses the singularity proliferation and restores a more uniform mapping, thereby stabilizing the skyrmion topology (d–i). Coherence-enhanced stability of the skyrmion number in strong atmospheric turbulence and comparison with total OAM. a–c Skyrmion number Nsk evaluated at the receiver plane for 200 independent turbulence realizations for a a fully coherent beam, b a partially coherent beam with δ0 = 1.0 mm, and c a partially coherent beam with δ0 = 0.5 mm. The mean value N¯sk and standard deviation σ are annotated in each panel. d–f Corresponding statistics of the total orbital angular momentum (OAM) per photon for the same turbulence realizations and coherence widths; the mean h¯ and standard deviation σ are shown. All panels correspond to strong turbulence with Cn2=1×10−8m−2/3. The scintillation indices (SI) of the received intensity for the three different coherence widths are highlighted in the figure. Strikingly, reducing the spatial coherence of the incident beam markedly enhances the topological robustness in turbulence. Although the Stokes-parameter patterns are still distorted by turbulence (see Supplementary Fig. 8), both the proliferation of phase singularities in S¯c and the irregular mapping on the parameter sphere are progressively suppressed when the coherence width becomes comparable to or smaller than the beam waist, as shown in Fig. 4d–i. This suppression mitigates the turbulence-induced degradation of Nsk. The dependence of the stability of Nsk on both the spatial coherence width and the beam waist is further demonstrated in Supplementary Sec. S5. Consistently, Fig. 5b, c show that partially coherent skyrmion beams exhibit a highly stable skyrmion number across different turbulence realizations. We note that Cn2=1×10−8m−2/3 corresponds to the strong-fluctuation regime, as indicated by the intensity scintillation index (SI) of the fully coherent case [SI = 1.3223 > 1; Fig. 5a]. Although SI decreases with reduced coherence [Fig. 5b, c], it remains above unity even for δ0 = 0.5 mm, confirming that the observed stabilization of Nsk is not a consequence of entering a moderate or weak-turbulence regime. The underlying mechanism can be understood as follows. A partially coherent beam can be viewed as an incoherent superposition of many spatial modes (or, equivalently, a statistical ensemble of fields) within a single turbulence realization33,45. This modal incoherent averaging suppresses the formation and survival of singularities in S¯c and smooths the parameter-sphere mapping of S¯(r), thereby stabilizing the skyrmion number. In principle, a similar averaging could be achieved by time-averaging over many independent turbulence realizations29, but such an approach would require long acquisition times and is therefore impractical for applications that rely on rapid readout of skyrmion topology. In contrast, partial coherence can be generated and modulated rapidly, enabling real-time production of turbulence-resilient skyrmion states of light. Finally, we emphasize that this coherence-enabled robustness of the skyrmion number is distinct from the behavior of the total orbital angular momentum (OAM). In our simulations, the total OAM exhibits pronounced turbulence-induced fluctuations regardless of the input spatial coherence [Fig. 5d–f], in clear contrast to the stabilized skyrmion topology observed for partially coherent beams. Topological transitions induced by optical coherence Beyond its advantage in mitigating turbulence-induced degradation, optical coherence provides an additional degree of freedom to actively reshape skyrmion textures and even induce topological transitions during propagation. We demonstrate this capability with two classes of engineered non-Gaussian coherence functions. We first consider a partially coherent skyrmion beam whose degree of spatial coherence takes a Laguerre–Gaussian (LG) correlation form46, where Ln0 is the generalized Laguerre polynomial of radial order n (zero azimuthal order) and δ0 sets the coherence width [see the spatial distribution in Fig. 6a]. Under this distinctive coherence structure, the Stokes-vector texture exhibits a pronounced propagation-induced topological transition [Fig. 6b–e]: starting from a Néel-type skyrmion at the source plane (z = 0), the texture evolves into a skyrmionium at an intermediate distance around z = 0.9f, where the skyrmion number approaches Nsk ≈ 0. Such a skyrmion–skyrmionium conversion is generally absent in fully coherent skyrmion beams under smooth phase perturbations47,48. The propagation distance at which the transformed topological texture becomes fully developed depends on the engineered spatial coherence. In general, lower spatial coherence accelerates the coherence-induced reshaping process, so the transition can occur at a shorter propagation distance. Coherence-driven topological transitions of partially coherent optical skyrmions enabled by non-Gaussian spatial correlations. a Amplitude ∣μ(r1, r2)∣ and phase arg[μ(r1,r2)] of a Laguerre--Gaussian (LG) degree of coherence. b–e Propagation-induced skyrmion-to-skyrmionium transition under the LG coherence: shown are the normalized Stokes textures S¯(r) at the source plane and at selected propagation distances during focusing (z = 0, 0.5f, 0.9f, and f). The extracted skyrmion number Nsk is indicated below each texture. Insets encode arg(S¯1+iS¯2) and S¯3 as indicated by the color bars. f Amplitude and phase of a Hermite--Gaussian (HG) degree of coherence. g–j Coherence-enabled skyrmion splitting and multiplication under the HG coherence: shown are the normalized Stokes textures S¯(r) at z = 0, 0.5f, 0.9f, and f, where the initial skyrmion texture evolves into four spatially separated skyrmionic parts, resulting in an increased magnitude of the total skyrmion number. At the focal plane, the skyrmionium consists of two nested Bloch-type skyrmions with opposite polarity: a central component with polarity +1 and an outer component with polarity −1. The experimentally evaluated skyrmion number at focus is Nsk = 0.2884, which is primarily due to the finite integration domain used in the measurement (restricted to regions with intensity above 2% of the peak value). Extending the integration to the full transverse plane yields a skyrmion number close to zero, consistent with the skyrmionium character. The physical origin of this extraordinary transition lies in the discontinuous phase structure encoded in the engineered coherence function. As shown in Fig. 6a, the phase of μ(r1, r2) contains a circular π-phase dislocation, implying that the source experiences an effective phase-discontinuous perturbation in a statistical, ensemble-averaged sense. During propagation, the engineered spatial coherence structure modifies the evolution of the polarization matrix and thereby reshapes the Stokes texture. Consequently, the associated topological transitions do not appear at the source plane, but emerge progressively during propagation. These coherence-induced structures become most clearly developed in the far field, which corresponds to the focal plane in our experiment, as shown in Fig. 6. This behavior is in sharp contrast to a deterministic π-phase step imprinted directly on a fully coherent field. As a second example, we engineer a Hermite–Gaussian (HG) degree of coherence46, where Hm and Hn are Hermite polynomials of orders m and n, respectively. The amplitude of this coherence function forms a rectangular spot array, while the phase alternates by π between neighboring spots [Fig. 6f], indicating a coherence phase landscape with rectangular-symmetry discontinuities. Under this coherence structure, the Stokes texture gradually splits into four identical, spatially separated skyrmion parts during propagation [Fig. 6g–j]. Correspondingly, the skyrmion number evolves from Nsk = − 0.8709 at the source plane to Nsk = − 3.7818 in the focal plane (far field), with each sub-beam contributing approximately −0.94. This topological multiplication originates from the rectangularly symmetric, phase-discontinuous coherence landscape, which reshapes the skyrmion texture through propagation. These two examples highlight that distinct coherence structures can drive qualitatively different topological transitions, thereby establishing spatial coherence engineering as a versatile route for controlling skyrmion states of light. The corresponding numerical results for the two types of partially coherent skyrmion beams with non-Gaussian correlation agree well with the experimental observations (Supplementary Fig. 5). The coherence-induced transitions reported here are also applicable to other partially coherent skyrmion beams; for example, analogous behavior is obtained for a beam composed of partially coherent LG00 and LG02 modes (Supplementary Fig. 6). Discussion In this work, we have investigated how partial spatial coherence reshapes, stabilizes, and controls optical skyrmions encoded in polarization textures during propagation in both free space and atmospheric turbulence. A central difficulty in extending skyrmion topology from ideal coherent beams to realistic stochastic fields is the emergence of propagation-induced, spatially inhomogeneous depolarization, which causes the conventional Stokes-vector mapping to collapse into the interior of the Poincaré sphere and obscures a well-defined topological characterization. To address this, we introduced a reshaped parameter-space representation based on a local renormalization of the Stokes vector, enabling a consistent definition of partially coherent skyrmion fields and their skyrmion number in the presence of nonuniform polarization degree. With this framework, we identified a distinct coherence-enabled regime in which the skyrmion number remains stable during free-space propagation, and we further demonstrated that reducing spatial coherence can markedly enhance the robustness of skyrmionic topology in atmospheric turbulence. Importantly, this stabilization persists even in the strong-fluctuation regime (scintillation index exceeding unity), where fully coherent skyrmions exhibit pronounced degradation and realization-to-realization instability. These findings suggest a viable route toward real-time, topology-encoded free-space photonic links, in which information is carried by the skyrmion number rather than by fragile field details. Beyond robustness, we show that spatial coherence engineering constitutes an independent and powerful control degree of freedom for topological transitions. By tailoring non-Gaussian coherence landscapes with phase discontinuities, we realize propagation-induced topological transitions, including skyrmion-to-skyrmionium conversion and controlled splitting into skyrmion lattices. Notably, these transformations require no additional polarization components along the propagation path; instead, the desired topology is programmed at the source through the coherence structure and emerges spontaneously through propagation. This coherence-programmable approach opens opportunities for reconfigurable topological beam shaping and multiplexing, with potential relevance to robust structured-light communications, topological metrology, and sensing in complex environments. More broadly, our study highlights an overarching message: partial coherence, ubiquitous in natural and practical optical fields, is not merely an imperfection to be mitigated but a resource that can protect and even engineer topological states of light. Extending skyrmions from pristine lasers to stochastic photonic fields therefore enriches topological photonics with the tools and insights of statistical optics, and points to coherence-engineered topological textures as a practical platform for resilient photonic functionalities in the real world. Methods Optical skyrmions in the Stokes-vector texture The normalized Stokes vector of an optical beam is defined as where sj(r) (j = 0, 1, 2, 3) are the Stokes parameters evaluated at the transverse position r. In terms of the local polarization matrix Φ(r) = E*(r)ET(r), the Stokes parameters can be written as where σ0 is the 2 × 2 identity matrix and σ1,2,3 are the Pauli matrices. The local degree of polarization, which quantifies the fraction of fully polarized component in the total field at r, is given by Following Ref. 11, the Stokes texture of an optical skyrmion can be conveniently parameterized as where α(θ) describes the azimuthal winding of the transverse Stokes components and β(r) determines the radial evolution from the beam center to the boundary of the skyrmion domain. With this representation, the skyrmion number can be factorized as Nsk = q m, where Here q is the polarity, determined by whether S3 points down (up) at the center and up (down) at the boundary (corresponding to q = +1 or q = −1), and m is the vorticity associated with the azimuthal winding of the transverse Stokes field S⊥ = [S1, S2]. By tailoring the polarization basis, spatial modes, and the relative phase γ, a variety of Stokes-vector topological textures can be constructed, including Bloch- and Néel-type skyrmions, anti-skyrmions, merons, and bimerons11. Experimental set-up and description The schematic of the experimental system for generating partially coherent skyrmion beams and measuring their Stokes parameters is shown in Fig. 7. A linearly polarized continuous-wave laser (λ = 532 nm) is expanded by a beam expander (BE, 20×) and then split by a 50:50 beam splitter (BS1). One output arm illuminates a phase-only spatial light modulator (SLM; 1920 × 1080 pixels, pixel pitch 8 μm, refresh rate 56 Hz). The SLM display is divided into two halves, each playing a time sequence of computer-generated holograms (CGHs) to generate random field realizations. Time averaging over these random realizations yields partially coherent Laguerre–Gaussian modes: LG00 from the left half and LG0l from the right half of the SLM screen. Details of the CGH synthesis are given in Sec Hologram design. Experimental set-up for generating partially coherent skyrmion beams and measuring their Stokes parameters. A linearly polarized laser is expanded by a beam expander (BE) and split by a 50:50 beam splitter (BS1). One arm illuminates a spatial light modulator (SLM) displaying time-varying computer-generated holograms (CGHs) on two halves of the screen to generate two partially coherent scalar fields Eleft and Eright sharing the same random-screen realization. The modulated beams are processed by a common-path 4f system (L1-L2) with a V-shaped aperture (VA) to select the desired diffraction orders, and their polarizations are converted by half-wave plates (HWPs). A Ronchi grating (RG) recombines the two components into a single partially coherent vector skyrmion source. Stokes polarimetry (SP; quarter-wave plate + linear polarizer) together with a CCD camera is used to record six polarization-resolved intensity images (IH, IV, ID, IA, IR, IL) for reconstructing the Stokes parameters. Lens L3 images the source plane onto the CCD (Arm 2), and a second measurement arm (Arm 3, L4) translates along z to acquire Stokes parameters at different propagation distances. After reflection from the SLM, the field enters a common-path 4f system formed by lenses L1 and L2 (identical focal lengths). A V-shaped aperture (VA) placed at the Fourier plane between L1 and L2 suppresses the zero order and unwanted diffraction orders, transmitting only the two desired first-order beams. The two beams then pass through independent polarization-control elements (HWP1 and HWP2), which convert their linear polarizations into right- and left-circular polarizations, respectively. A Ronchi grating (RG) placed after the 4f system spatially overlaps the two beams, producing a single partially coherent vector skyrmion beam at the RG output. In our experiment, 1000 random modes are used to synthesize the partially coherent skyrmion beam, and these modes are displayed cyclically on the SLM. The spatial distributions of these random modes control both the coherence and topological structures of the synthesized beam. We note that this SLM-based implementation is intended as a proof-of-principle demonstration of coherence-engineered skyrmion generation, rather than as a real-time source for practical turbulence communication experiments. In a real-time turbulence scenario, the source fluctuation time should be much shorter than the characteristic turbulence evolution time, which is typically on the scale of tens of milliseconds49. Therefore, the SLM used in our experiment is too slow for this specific real-time turbulence condition. A practical implementation could instead use a much faster spatial modulator, such as a digital micromirror device (DMD), which can operate at kHz to tens-of-kHz rates49,50. For example, with 1000 random patterns and a refresh rate in the tens-of-kHz range, the synthesis time can be reduced to the millisecond scale, well below the characteristic evolution time of atmospheric turbulence. Thus, while the present SLM experiment demonstrates the principle of coherence-engineered skyrmion generation, faster modulators would be required for real-time applications in dynamic turbulent channels. To characterize the skyrmion source plane (immediately after the RG), an imaging lens L3 relays the source plane onto a CCD camera with unit magnification. A polarization analyzer, consisting of a quarter-wave plate followed by a linear polarizer, is placed in front of the CCD to perform conventional Stokes polarimetry (SP) and retrieve s0–s3, from which the Stokes texture and skyrmion number are computed (see Sec Stokes polarimetry). To measure the propagation dynamics, the analyzer and CCD are mounted on a motorized translation stage and scanned along z, allowing acquisition of the Stokes parameters at different propagation distances. Hologram design To generate two partially coherent orthogonally polarized modes simultaneously, the SLM is divided into left and right halves that encode, respectively, the target deterministic fields Eleft(r) (LG00) and Eright(r) (LG0l). To realize the partially coherent vector beam consistent with Eq. (3), we impose the same (time-varying) random complex screen Tn(r) on both halves, where n = 1, 2, …, so that the two polarization components remain fully correlated within each realization. The SLM output remains the same linear polarization for two modes; the conversion to left- and right-circular polarizations is implemented downstream by HWP1 and HWP2. The random screens Tn(r) are synthesized assuming Gaussian statistics. Specifically, we generate complex Gaussian white noise Rn(f) (zero mean and unit variance) in the spatial-frequency domain and filter it by the square root of the desired power spectrum p(f), yielding where F−1 denotes the inverse Fourier transform. The correlation of the screens satisfies μ(r1,r2)∝⟨Tn*(r1)Tn(r2)⟩, and thus determines the prescribed degree of coherence. The spatial coherence structure, thus, is controlled by the power spectrum p(f). For each realization n, the complex fields Eleft,n(r) and Eright,n(r) are encoded into a phase-only CGH following the method in ref. 51. The resulting CGH sequence is displayed cyclically on the SLM to realize a temporally varying ensemble of vector-field realizations. Stokes polarimetry The Stokes texture, skyrmion density, and skyrmion number are reconstructed from experimentally measured Stokes parameters. We perform conventional Stokes polarimetry (SP) using a quarter-wave plate (QWP) and a linear polarizer placed in front of the CCD. By setting the QWP and polarizer at appropriate angles, we record the time-averaged intensity profiles in six polarization channels: horizontal (IH), vertical (IV), diagonal (ID), anti-diagonal (IA), right-circular (IR), and left-circular (IL). The Stokes parameters are then obtained as Representative measurements of the six polarization-resolved intensity profiles for a partially coherent skyrmion beam composed of LG00 and LG01 spatial modes are shown in the inset of Fig. 7. Supplementary information Supplementary Information Transparent Peer Review file Acknowledgements The authors thank the reviewers for their constructive comments and valuable suggestions. Author contributions F.W., Y.S., Y.H.C., and Y.L. conceived the idea. The design, theoretical analysis and numerical simulation were carried out by Y.L., and F.W., and were verified in experiment by Y.L. Y.L. collected the data; Y.L., Z.G., S.C., K.Z., Y.H.C., Y.J.C., Y.S., and F.W. carried out the data analysis. Y.L., Y.H.C., and F.W. contributed to writing–original draft and all authors contributed to the writing–review and editing. The entire research was supervised by Y.H.C., Y.S., Y.J.C., and F.W. All authors have read and agreed to the published version of the manuscript. Peer review Peer review information Nature Communications thanks Jiawen Chen, Yuhan Dong, Sheng Liu and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding Y.J.C. discloses support for the research of this work from the National Natural Science Foundation of China (grants 12192254, 12534014, W2441005), the National Key Research and Development Program of China (2022YFA1404800), the Natural Science Foundation of Shandong Province (ZR2025ZD21), and the Key Research and Development Program of Shandong Province (2024JMRH0105). Y.L. discloses support for the research of this work from the National Natural Science Foundation of China (12404348). Y.H.C. discloses support for the research of this work from the National Natural Science Foundation of China (12274310). F.W. discloses support for the research of this work from the National Natural Science Foundation of China (12274311). Y.S. discloses support for the research of this work from the Singapore Ministry of Education (MOE) AcRF Tier 1 (RG157/23 & RT11/23), the Singapore Agency for Science, Technology and Research (A*STAR) (M24N7c0080 & H25-MRO3489), and the Nanyang Assistant Professorship Start Up grant. The other authors (S.C., Z.G., and K.Z.) declare no relevant funding. Data availability The data supporting the findings of this study are available within the paper and its Supplementary Information. The corresponding authors can also provide the data upon request. Code availability The code used to produce the results is available from the corresponding authors upon request. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Yahong Chen, Email: yahongchen@suda.edu.cn. Yangjian Cai, Email: yangjian_cai@163.com. Yijie Shen, Email: yijie.shen@ntu.edu.sg. Fei Wang, Email: fwang@suda.edu.cn. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77021-1. References Associated Data Supplementary Materials Supplementary Information Transparent Peer Review file Data Availability Statement The data supporting the findings of this study are available within the paper and its Supplementary Information. The corresponding authors can also provide the data upon request. The code used to produce the results is available from the corresponding authors upon request.