Over the last decade, the study of topological phases of matter has made fruitful achievements with far-reaching impacts. A particular example is the discovery and characterization of hundreds and thousands of topological materials protected by time-reversal and/or crystal symmetries1–8. Symmetry plays an essential role in distinguishing topologically inequivalent band structures3,9; this has recently culminated in systematic classification schemes that address the global band topology in terms of irreducible representation combinatorics and symmetry-based indicators10–15. The paradigm that has led to this success of topological band theory is to divide energy band structure into two subspaces, i.e., an “occupied” subspace spanned by bands below an energy gap and an “unoccupied” subspace above the gap. This yields additive group structures for topological invariants when adding extra occupied bands. Physical systems captured by this framework of “single-gap” topology are often characterized by abelian groups with commutativity and additivity, with a prominent example being the tenfold way classification rooted in the *K*-theory16–18.

Very recently, symmetry-protected topological phases beyond the aforementioned abelian classifications have been predicted theoretically. With multiple bandgaps entangled together, the underlying new paradigm is characterized by non-abelian groups19–22. The multi-gap topology exhibits appealing noncommutative braiding structures and reflects an entirely new category of band topology. The braiding in energy-momentum space can be described by non-abelian charges20–22. For example, a \(PT\) or \({C}_{2z}T\) symmetric three-band system can host a real-valued Hamiltonian and the three real eigenstates in Hilbert space constitute an orthonormal frame20 − 23, and the eigenstate trajectories along a closed path define a non-abelian frame charge for the system. According to the homotopy group theory, the charge is associated to the flag variety \(\text{S}\text{O}\left(3\right)/{\text{D}}_{2}\) and described by anticommutative quaternion group20. Particularly, the multi-gap nodes can braid reciprocally in two-dimensional (2D) momentum space as they evolve with external parameters. The non-abelian quaternion charge of a node can change sign through braiding with another node in an adjacent gap, which can result in a pair of nodes of an identical charge in the same gap. The identically charged nodes will not annihilate after collision, as the key signature of a nontrivial nodal braiding process20 − 22. Besides the quaternion charges, the nodal braiding can also be characterized by Euler classes21,24 and Dirac strings19,25.

Non-abelian topological systems have been ingeniously realized by photonic26–28, transmission line network27,29 and acoustic30,31 means. And relevant phase transitions have also been clearly observed30. However, unambiguous experimental evidence for the long-desired nodal braiding remains elusive, because of the challenges in resolving the evolution of nodal points in energy-momentum space, demanding the extraction of wavefunction information in Hilbert space, and hence concluding the non-abelian phase transition at the ultimate level of wavefunctions. Here, we report an elegant acoustic realization of non-abelian nodal braiding using a simple square-lattice model. Unambiguously, we observe a nontrivial nodal braiding process in energy-momentum space, benefiting from the extremely clean bulk dispersions with minimal nodes residing only at high symmetry lines (HSLs) of the Brillouin zone (BZ). Particularly, we present the first experimental evidence, at the level of wavefunctions, for the band inversion and the resultant stability transition of a colliding nodal pair at a high symmetry point (HSP), as a hallmark manifesting the nontrivial multi-gap braiding effect. Moreover, we explore the intricate bulk-edge physics of our 2D system from the perspective of braiding 1D subsystems, and unveil the nodal braiding through a global evolution of multi-gap edge states.

**Minimal model for non-abelian nodal braiding**

Figure 1a illustrates two pairs of band nodes in a 2D three-band system with \({C}_{2z}T\) symmetry. Each node carries a non-abelian charge20 − 22 (\(Q\)) in the anticommutative quaternion group \({G}_{Q}=\{\pm 1,\pm i,\pm j,\pm k\}\), in which the elements satisfy the fundamental multiplication rules \({i}^{2}={j}^{2}={k}^{2}=-1\) and \(ij=k,jk=i,ki=j\). Upon introducing time-evolved system parameters, the nodes can move in momentum space and even collide somewhere (usually at a HSP). The total charge of two colliding nodes in the same bandgap is determined by their product. If the result is \(+1(-1)\), the band nodes will (will not) annihilate pairwise after collision20−22. Here we consider a nontrivial braiding process that transfers an unstable nodal pair \((Q=+1)\) to be a stable one \((Q=-1)\), as sketched in Fig. 1b by the time-evolved trajectories of the gap-I nodes (blue lines) and gap-II nodes (red lines). In general, a pair of newly created band nodes (green stars) carry opposite quaternion-valued frame charges (QFCs) and their total charge is \(+1\). The charge sign of a given node can be flipped via braiding with another node of an adjacent gap, as described by the multiplication rule of QFCs. For example, after braiding with the gap-I node \(\text{A}\) (\(Q=-k\)), the gap-II node \(\text{B}\)changes its charge (at the red triangle) from \(-i\) to \(+i\), i.e., \(\left(-k\right)\cdot \left(-i\right)\cdot {\left(-k\right)}^{-1}=+i\)20,22. Subsequently, the identically charged gap-II nodes \(\text{B}\) and \({\text{B}}^{{\prime }}\) (\(Q=+i\)) cannot annihilate even after collision, and instead they bounce to another momentum line (Fig. 1b). Thus, the collision stability reflects the topology of a nodal pair in the same gap, and the topological phase transition can be induced by a nontrivial multi-gap nodal braiding.

We use a simple three-band square lattice model (Fig. 1c) to realize this non-abelian braiding process, which involves only one pair of band nodes in each gap21. The Hamiltonian of this model reads

where the hoppings and . The onsite energies of the orbitals 2 and 3 are , the hopping strength between the orbitals 1 and 2 is , whereas the onsite energy of the orbital 1, , and the intra- and inter-cell hoppings between the orbitals 2 and 3, and , vary with time, whose functional relations are given in Fig. 1d. The system features mirror symmetries and , which protect the nodal point degeneracy in the () and () diagonals of the square BZ. The Hamiltonian can be made real-valued at all momenta** **after a** **unitary transform, thanks to the symmetry since (*Supplementary Information 1*).

Based on the tight-binding model, we have calculated the full-BZ band structures at a series of representative moments (Extended Data Fig. 1). An example can be seen in Fig. 1a (\(t=6\)), which features gap-I nodes \(\text{A} \left({\text{A}}^{{\prime }}\right)\) and gap-II nodes \(\text{B} \left({\text{B}}^{{\prime }}\right)\). Figure 1e shows the time evolution of their QFCs, calculated with a consistent gauge selection (*Supplementary Information 2*). Figure 1f sketches the nodal positions and their subsequent movements in the square BZ. At the initial time (\(t=0\)), the three bands are energetically separated and host no band nodes. Next, a time-reversal-related nodal pair \(\text{A} \left({\text{A}}^{{\prime }}\right)\) of \(Q=\pm k\) emerge at \({\Gamma }\)-point (\(t=1\)) and move in the (\(1\stackrel{-}{1}\)) direction (\(t=2\)). Similarly, at \(t=4\) and \(6\), another nodal pair \(\text{B} \left({\text{B}}^{{\prime }}\right)\) of \(Q=\pm i\) nucleate at \(\text{M}\)-point and move in the (\(11\)) direction. The two pairs of nodes are braided reciprocally, resulting in sign-flipped QFCs for the nodes \({\text{A}}^{{\prime }}\)and \(\text{B}\) (\(t=7\)). The identically charged nodal pairs will not annihilate after their collision, as exemplified by the gap-II nodes \(\text{B}\) and \({\text{B}}^{{\prime }}\) (see \(t=7\tilde10\)). The collision stability of the nodal pair can also be characterized by a nontrivial Euler class defined over a patch around \({\Gamma }\)-point (\(t=8\)), in contrast to the trivial Euler class defined around \(\text{M}\)-point at its creation (\(t=4\)). The (two-band) calculation of Euler class is straightforward and omitted here21. For brevity, hereafter we use the QFC to describe the whole three-band braiding physics.

**Ideal acoustic braiding metamaterials**

We have designed ideal acoustic metamaterials to realize the introduced braiding effect. Figure 2a shows a photo of our acoustic metamaterial. As sketched in Fig. 2b, each unit cell contains three air cavities connected by narrow tubes. Physically, the cavity resonators emulate atomic orbitals, and the narrow tubes mimic hoppings between them32-34. We consider the lowest dipole resonance mode polarized along the length direction of each cavity. In particular, the structural parameters , , and simulate the three variable parameters of the braiding Hamiltonian , i.e., , , and , respectively. Given that the resonance frequency is inversely proportional to the cavity length and that the coupling strength is (roughly) proportional to the cross sectional area of the narrow tube, we present the time dependences of , , and in Fig. 2c, which are piecewise functions like Fig. 1d. (See details in *Supplementary information 3*.) Figure 2d shows the simulated acoustic band structures at eight typical moments, which describes a nodal braiding equivalent to Fig. 1f. Excellent consistency is attained between the full-wave simulations and the model results (see Extended Data Fig. 2). Note that different from the bilinear dispersion around a general HSL node, the unique biquadratic (hybrid) line shape around a stable (unstable) HSP node can be directly used to visualize the collision stability of a nodal pair (see *Supplementary Information 4*).

The non-abelian nodal braiding unavoidably involves multi-band wavefunction entanglements. Below we elucidate that one can identify the stability of a colliding nodal pair through detecting the mirror eigenvalues of HSP Bloch wavefunctions. Specifically, along the HSL (\(1\stackrel{-}{1}\)) or (\(11\)) two bands of opposite mirror eigenvalues can cross to form a pair of time-reversal-related nontrivial nodes7,21, which may collide at the HSP \({\Gamma }\) or \(\text{M}\) as the time evolution. Therefore, the collision stability of a nodal pair can be inspected from the mirror eigenvalues at given HSP: the nodal pair will not annihilate after collision if the eigenvalues of both mirrors are opposite for the two crossing bands; otherwise, the nodes will annihilate pairwise. To show this, we label the mirror eigenvalues of the HSP states on the band structures of \(t=0\), 6 and 10. As exemplified by the moment \(t=6\), the second and third bands carry opposite eigenvalues for both mirrors at \({\Gamma }\), which means that the colliding nodal pair in gap-II will not annihilate at \({\Gamma }\), but bounce from (\(11\)) to (\(1\stackrel{-}{1}\)) direction (see next moments in Figs. 1f and 2d); on the contrary, the colliding nodes would pairwise annihilate at \(\text{M}\) since only the eigenvalues of \({\mathcal{M}}_{1\stackrel{-}{1}}\) are opposite there, which coincides with their pairwise creation at \(t=4\) if we reverse the time evolution from \(t=6\) to \(t=2\). Similarly, one can predict the collision stability of the gap-I nodes: unstable at \({\Gamma }\) (see \(t=0\tilde2\)) but stable at \(\text{M}\) (after \(t=10\), not shown here). To visualize the HSPs’ mirror eigenvalues, we present the eigenfield profiles in Fig. 2e for the moment \(t=6\). Intriguingly, the pressure patterns exhibit a nature of sublattice polarization, i.e., the orbital 1 and the other two orbitals cannot be occupied simultaneously, as enforced by the mirror symmetries of the system.

**Experimental characterization of acoustic nodal braiding**

Our acoustic metamaterials were fabricated with using 3D printing technology. Each sample consists of \(15\times 15\)unitcells, and the total size is ~\(60.0 \text{c}\text{m}\times 60.0 \text{c}\text{m}\). To excite and scan the pressure field, the cavities are perforated with small holes (sealed when not in use) for inserting the sound source and detector. We place the sound source in the middle of the sample, and scan the real-space pressure distribution cavity by cavity. Finally, band structures in momentum space were achieved by performing 2D spatial Fourier transform (see details in *Supplementary Information 5* and Extended Data Fig. 3).

Figure 3a shows the band structures (bright color) measured along the HSLs (\(11\)) and (\(1\stackrel{-}{1}\)) of the square BZ. The experimental data, well-resolved in energy-momentum space, perfectly reproduce our numerical results (white dashed lines) except for the band broadening due to the dissipation and finite-size effect. In particular, our experimental data resolve clearly a biquadratic line shape around the stable gap-II node at \({\Gamma }\) (\(t=8\)), in contrast to the hybrid line shape sprouted from an unstable HSP node, i.e., the gap-I node created at \({\Gamma }\) (\(t=1\)) or the gap-II node created at \(\text{M}\) (\(t=4\)), where the linear crossing happens along the HSL (\(11\)) or (\(1\stackrel{-}{1}\)). For comparison, bilinear line shapes were checked for the nontrivial nodes at general HSL momenta (see Extended Data Fig. 4). Furthermore, in Fig. 3b we present the 2D spatial Fourier maps at the nodal frequencies for the instants \(t=4\tilde10\), which directly visualize the creation, moving, collision and bounce of the gap-II nodes \(\text{B}\left({\text{B}}^{{\prime }}\right)\) in momentum space. This is another experimental evidence for the stable node (\(t=8\)) evolved from the unstable one (\(t=4\)) after braiding with the gap-I nodes \(\text{A}\left({\text{A}}^{{\prime }}\right)\). Thus far, we experimentally characterized a concise but nontrivial nodal braiding process, from a level of Fourier spectra.

The mirror eigenvalues of the HSP states, which can be used to conclude the stability of HSP nodes, were confirmed in our acoustic experiments. We focus on the case of first. Specifically, we extracted the normalized wavefunctions at the HSPs from the measured 2D Fourier spectra, (with ), and calculated their mirror expectation values, (see *Supplementary Information 5*). Theoretically, approaching points to the mirror symmetric (antisymmetric) state of eigenvalue . Figure 4a provides the measured expectation value spectra for the HSPs and . As expected, for both cases, the spectra are peaked toward or dipped toward near each predicted eigenfrequency (dashed line), around which the sound intensity is maximized (see Extended Data Fig. 5). The mirror symmetries of the HSP states can also be directly inspected by the measured wavefunctions in Fig. 4b. Remarkably, the wavefunctions exhibit almost vanished pressure fields as predicted in Fig. 2e, which verifies the nature of sublattice polarization experimentally. Similar measurements were performed for the initial and final moments (see Extended Data Fig. 5). In particular, figures 4c and 4d present the mirror expectation values extracted for the three bands at and . Consistent with the mirror eigenvalues depicted in Fig. 2d (see and ), our experimental data clearly unveil the band inversions induced by multi-gap nodal braiding, which are inevitable intermediate steps to change the collision stability of nodal pairs. A more complete experimental characterization for the band inversion can be seen in Extended Data Fig. 6. The above experiments present evidence, from a deeper wavefunction level, for the band inversion and the consequent creation of stable HSP node, as the crucial signature of a nontrivial nodal braiding.

**Edge manifestation of non-abelian nodal braiding**

Although the theme here is nodal braiding in 2D bulk, we point out that our \({C}_{2z}T\) symmetric system can also be used to study 1D braiding physics if considering its subsystems formed by \({k}_{y\left(x\right)}\)-loops, where \({k}_{x\left(y\right)}\) serves as a braiding parameter for each 2D sample. Here we focus on the 1D subsystem of constant-\({k}_{x}\), whose non-abelian topology can be characterized by the QFC defined for given \({k}_{y}\)-loop, \(q\left({k}_{x}\right)\). Interestingly, as exemplified with \(t=2\) (Fig. 5a), the loop charge can be related to the nodal charge by a simple quotient relation \(Q={q}_{r}/{q}_{l}\), where \({q}_{r}\) and \({q}_{l}\) describe the \({k}_{y}\)-loops at the right and left sides of the node (see *Supplementary Information 2*). As such, the 1D non-abelian topological transition (where \(q\) changes) occurs when meeting a nontrivial node in the evolution of \({k}_{x}\). This is illustrated in Fig. 5b by the \({k}_{x}\)-dependent loop charge, which jumps at the nodal projections. More importantly, the loop charge is closely related to the quantized Zak phases of the three gapped bands at given \({k}_{x}\), and thus can be used to predict the presence of edge states in both gaps27, as summarized in Fig. 5c. (The case of \(q=-1\) is an exception that goes beyond the Zak phase description27, which does not happen in our system.) Therefore, we can explore the bulk-edge physics of each 2D sample from the perspective of braiding 1D subsystems, and reflect the bulk nodal braiding through the global evolution of multi-gap edge states.

The multi-gap bulk-edge physics was convincingly evidenced by our acoustic experiments (see details in *Supplementary Information 5*). Figures 5d-5k show the measured edge spectra (color scale). All data capture well the simulations (color lines), benefiting again from the minimal nodal pairs in our braiding model. Notice that the edge responses are fully consistent with the theoretical predictions summarized in Fig. 5c, according to the \({k}_{x}\)-dependent loop charges labeled in Figs. 5d-5k. Specifically, for the moment of \(t=0\), Fig. 5d exhibits no edge states because \(q=+1\) everywhere; a similar result appears in Fig. 5e (\(t=1\)), except for the closure of gap-I at \({k}_{x}=0\) associate to the creation of a trivial node (\(Q=+1\)). For the moments of \(t=2\) and \(4\), Figs. 5f and 5g show clearly topological edge states connecting the projected gap-I nodes, because the loop charge in this \({k}_{x}\) interval becomes \(q=+k\) thanks to the split nontrivial nodes \(\text{A}\) and \({\text{A}}^{{\prime }}\) (\(Q=\pm k\)). With the emergence of the gap-II nodes \(\text{B}\) and \({\text{B}}^{{\prime }}\) (\(t=6\)), new edge states appear in this gap (Fig. 5h), consistent with the nontrivial loop charge \(q=-i\) for the corresponding momentum range. Besides, there are no edge states appearing in the \({k}_{x}\) intervals between the gap-I and gap-II nodal projections, as a reflection of the loop charge \(q=+1\). These facts conclude that the gap-II nodes belong to the conjugacy class \(\left\{\pm i\right\}.\) For the following moments (Figs. 5i-5k), new loop charges \(q=\pm j\) emerge between the gap-I and gap-II nodal projections, so that both gaps host edge states in these momentum intervals. More importantly, from the time evolution of gap-II edge states, one may infer the emergence of a stable node (\(Q=-1\)) at \(t=8\): there are edge states emitted from the pairwise nontrivial nodes before and after their collision. In contrast, no edge state appears before the creation of an unstable node (\(Q=+1\)). All the above edge manifestations reflect faithfully the physics of the non-abelian braiding of bulk nodes.