{"id":"3c96d14e-7775-4f1b-a7b5-b11a34ca7bb3","arxiv_id":"2411.18194","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"mumax+ is a new extensible, Python-interfaced, GPU-accelerated finite-difference micromagnetic solver that adds antiferromagnet, ferrimagnet, and magnetoelastic capabilities to the standard mumax3 feature set.","lead":"This paper introduces mumax+, an open-source GPU-accelerated simulator for magnetic materials, wrapped in Python and designed to be easy to extend. It demonstrates the code on three problems: current-driven domain wall motion in a polycrystalline antiferromagnet, a strain-driven racetrack memory in nickel oxide, and stray-field imaging of the non-collinear antiferromagnet Mn3Sn.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (5) boundary condition omits intersublattice DMI terms that follow from the DMI energy Eq. (4), so the AFM solver's surface behavior is not justified for the general advertised capability.","rationale":"The reader's verdict was CONDITIONAL with the AFM exchange/BC premise as the weakest assumption. I agree with that assessment but sharpen it: the most concrete defect is that Eq. (5) does not follow from the DMI energy Eq. (4) for intersublattice couplings. This is a correctness risk, not merely a parameterization issue. The demos likely avoid the defective branch, so the reported behaviors may still be genuine outputs; the paper's advertised generality (a third DMI tensor for intersublattice interaction) is what is not supported. A two-step test (analytic variation plus a targeted numerical benchmark against an atomistic model) would settle it. The critique is on the argument, not the authors; the code is open-source and the standard benchmarks are deferred to the repository, so a targeted benchmark is the right check.","tokens_in":13610,"tokens_out":17918,"duration_ms":154633,"concrete_test":"1) Analytically vary the full energy (Eqs. 3 and 4) and write the exact boundary condition; check whether it reduces to Eq. (5) only when D^{(s,s')}=0 for s≠s'. 2) If the missing terms are nonzero, run a minimal mumax+ test: a two-sublattice AFM slab with only intersublattice DMI (D^{(1,2)}≠0, D^{(1,1)}=D^{(2,2)}=0), free surfaces, relax from a random initial state, and compare the equilibrium surface spin configuration with an independent atomistic spin-lattice simulation using the same exchange and DMI parameters. Any systematic deviation at the surfaces confirms the boundary condition is misimplemented and requires a code fix or corrected documentation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.1.3 defines the general DMI energy as E = Σ_{s,s'} D^{(s,s')}_{ijk} m_j^{(s)} ∂_i m_k^{(s')} (Eq. 4) and advertises a third tensor for intersublattice DMI. The Neumann condition Eq. (5) is then asserted for each sublattice. Varying Eq. (4) with respect to m^{(s)} yields, after integration by parts, a surface term n_i D^{(s,s')}_{ijk} m_j^{(s')} δm_k^{(s)} summed over all partner sublattices s'. For s' ≠ s this term is proportional to m^{(s')}, not m^{(s)}. Equation (5), however, contains only \\hat{D}^{(s)} · (n ⊗ m^{(s)}), i.e. a single-sublattice DMI contribution; the intersublattice boundary terms are absent. Hence, if Eq. (5) is implemented as written, the code does not solve the boundary-value problem for the general DMI+exchange energy in antiferromagnets. This is load-bearing because the paper's central claim is a general finite-difference solver for AFMs with DMI. The three demonstrations use interfacial/intralattice DMI and local homogeneous DMI, so they may not expose the bug, but the advertised capability for non-collinear AFMs with intersublattice gradient DMI is unsupported. The reader's weakest_assumption correctly identified Eq. (5) as fragile but did not name this specific missing term.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents mumax+, an open-source GPU-accelerated finite-difference micromagnetics package with a Python interface, positioned as an extensible successor to mumax3. The reported advances are the ability to simulate antiferromagnets and ferrimagnets with multiple sublattices, a generalized DMI description, and coupled elastodynamics with magnetoelasticity. Three demonstrations are used to support these claims: current-driven domain-wall mobility in a polycrystalline antiferromagnet, a strain-driven domain-wall racetrack in NiO, and stray-field images of non-collinear Mn3Sn compared with NV magnetometry. The Methods section defines the LLG equation, exchange fields, magnetostatic fields, DMI energy and boundary conditions, elastic wave equation, and magnetoelastic couplings.","tokens_in":13840,"tokens_out":9545,"duration_ms":82663,"significance":"If the implementation is correct, mumax+ would be a valuable community resource: it is open-source, GPU-accelerated, and Python-based, and the paper explicitly ships a repository with tests. The paper contains some genuine external anchors: the 0% grain-boundary mobility curve agrees with the analytical model of Ref. [31], and the Mn3Sn calculation is compared to published NV images. However, the correctness of the advertised general DMI capability is called into question by the boundary condition in Eq. (5), and the Mn3Sn demonstration uses parameters whose provenance and sensitivity are not documented. These issues do not necessarily invalidate the three demonstrations, but they do weaken the broad claim that mumax+ is a general solver for non-collinear antiferromagnets with DMI.","major_comments":[{"comment":"Equation (5) contains only the intrasublattice DMI boundary term D^(s) · (n ⊗ m^(s)). Varying the general DMI energy of Eq. (4) with respect to m^(s) produces, after integration by parts, surface terms n_i D_{ijk}^{(s,s')} m_j^(s') δm_k^(s) summed over partner sublattices s'. For s' ≠ s these terms are proportional to the partner magnetization and are absent from Eq. (5). Thus the advertised 'third DMI tensor' for intersublattice interactions does not enter the boundary-value problem as written, and the code does not solve the general DMI+exchange variational problem for antiferromagnets. The three demonstrations use intralattice/interfacial and homogeneous DMI only, so they do not expose the omission. The paper should either derive and implement the full boundary condition, or explicitly restrict the DMI capability claim to intrasublattice and homogeneous DMI.","section":"3.1.3, Eq. (5)"},{"comment":"The exchange constants (10, -15, -25 pJ/m) and the homogeneous DMI (10 MJ/m^3) in the Mn3Sn simulation are presented without citation, and the agreement with Fig. 3(b) is only visual. The claim to 'reproduce experimentally observed domain structures' therefore needs a statement of whether these parameters are literature values or were adjusted, together with a sensitivity check (e.g., how much the pattern changes when the intergrain exchange reduction or the DMI strength is varied). Without this, the apparent agreement could be a consequence of parameter choice rather than a validated prediction of the model.","section":"Fig. 3 caption and Section 2.3"},{"comment":"The paper asserts that all capabilities have been thoroughly tested against the standard micromagnetic problems, mumax3, and analytical results, but no test outcomes are reported and the validation is deferred to an external repository. Since the central claim of the paper is that the solver is correct, the manuscript should include at least a concise summary of the verification tests (which problems were run, what error metrics were used, and the outcomes), even if the full test scripts remain online.","section":"Section 2 and Section 4"}],"minor_comments":[{"comment":"There are small typographical errors: 'magnetoelatic' should be 'magnetoelastic' in Table 1, and Eq. (7) reads 'andftot' with a missing space.","section":"Table 1 and Eq. (7)"},{"comment":"Several references are incomplete or lack journal names, e.g., Refs. [26], [42], [43], [44], [45], [46], [47], and [48] have volume/page or DOI information but no journal title; please restore the full bibliographic metadata.","section":"References"},{"comment":"After Eq. (6), the phrase 'The latter lies parallel to the principal axis of the material' is ambiguous; it would be clearer to state that the implemented homogeneous DMI vector is restricted to that direction or that an arbitrary d-vector can be set in the input.","section":"Section 3.1.3"},{"comment":"The sentence in Table 1 that the interfacial DMI 'is chosen to stabilize the domain wall' is honest but it would help to state whether the two domain walls are static and pinned in the absence of strain over the simulated time; this would establish that the motion is caused by the elastic pulse rather than by spontaneous relaxation.","section":"Section 2.2 caption to Table 1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is suitable for the journal if the boundary-condition issue is resolved or the DMI claim is narrowed, and if the Mn3Sn parameter provenance is documented. The missing boundary terms are not exercised by any of the three demonstrations, so the demonstrations themselves may stand, but the general-solver claim in the abstract is broader than what Eq. (5) supports."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nYou should know: this is a real software paper with a genuine deliverable. mumax+ is a from-scratch GPU micromagnetics code with a Python interface, supporting multiple magnet objects, tensor DMI, antiferromagnets with magnetoelastic coupling, and several Runge-Kutta integrators. The three demonstrations are plausible and reference external benchmarks: the polycrystalline AFM mobility curve matches the analytical model at 0% grain reduction, the NiO racetrack works with disclosed parameters, and the Mn3Sn stray-field patterns look qualitatively like the NV images. The writing is clear and the methods section gives enough detail to implement most of it.\n\nNow the soft spots. I agree with the reader's two concerns: the full test suite is deferred to an unpinned GitHub repo, so a referee can't verify the claimed standard-problem pass rates; and the Mn3Sn parameters (exchange constants, homogeneous DMI) are not cited, so the 'agreement' is more suggestive than rigorous.\n\nBut there's a more specific technical issue. Equation (5) gives the Neumann boundary condition for each sublattice s, but it only contains the intralattice DMI term \\hat{D}^{(s)} · (n ⊗ m^{(s)}). The general DMI energy in Eq. (4) includes intersublattice coupling D^{(s,s')} with s≠s'. Varying Eq. (4) with respect to m^{(s)} yields surface terms proportional to n_i D^{(s,s')}_{ijk} m_j^{(s')} — that is, a dependence on the partner sublattice. Those terms are missing from Eq. (5). So as written, the boundary condition is incomplete for the advertised general intersublattice DMI. The three demonstrations don't seem to exercise this — they use intralattice interfacial DMI or the homogeneous DMI of Eq. (6) — so the central results likely stand, but the claim that mumax+ is a general AFM solver with tensor DMI is not yet supported.\n\nThis is fixable: the authors should either implement the full boundary terms or explicitly restrict the feature to intralattice DMI. A referee should request that, and ideally a simple test with nonzero D^{(1,2)} to show no torque leaks at the boundary.\n\nOverall: this paper deserves serious peer review. It's a useful tool for the micromagnetics community, and the DMI issue should be caught before publication. I'd bring it to a reading group and I'd cite it once the boundary condition is clarified. Send it out.\n\nBest,","headline":"Real software, real value, and one real hole in the advertised AFM DMI boundary conditions that the authors need to close.","tokens_in":14530,"tokens_out":4478,"would_cite":true,"duration_ms":36177,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper presents mumax+, a GPU finite-difference solver that extends micromagnetics to antiferromagnets, non-collinear order, and magnetoelasticity, and demonstrates it on domain-wall, racetrack, and Mn3Sn simulations.","keywords":["antiferromagnetism","micromagnetics","finite-difference method","GPU acceleration","magnetoelastic coupling","domain-wall motion","non-collinear antiferromagnet","racetrack memory"],"falsifier":"Send a single calibrated 8 GHz shear traction pulse through a NiO strip with notches at 300 nm and 600 nm: the model predicts each domain wall moves exactly one notch spacing, 150 nm, so if a single pulse moves the walls by zero or by more than one notch, the coupled magnetoelastic implementation is wrong.","tokens_in":13313,"feed_emoji":"🧲","tokens_out":9433,"duration_ms":76050,"temperature":0.7,"pith_summary":"The paper presents mumax+, a GPU-accelerated micromagnetic simulator with a Python interface, and argues that one finite-difference solver can handle ferromagnets, collinear and non-collinear antiferromagnets, and ferrimagnets by resolving multiple sublattices. Its main new physics is full magnetoelastic coupling for antiferromagnets, joining magnetization dynamics to an elastodynamic equation whose strain feeds back through the inverse magnetostrictive, or Villari, effect. To show this works, the authors simulate three current problems: current-driven domain-wall motion in a polycrystalline antiferromagnet, a strain-pulse-driven antiferromagnetic racetrack memory in NiO, and the stray-field patterns of Mn3Sn as seen by nitrogen-vacancy magnetometry. If correct, the package gives the community a common platform for testing spintronic and straintronic proposals in systems where ferromagnetic approximations fail.","feed_headline":"GPU solver simulates antiferromagnetic racetrack memory","feed_subtitle":"Mumax+ reproduces domain-wall motion, a NiO strain-driven racetrack, and Mn3Sn NV images.","key_machinery":"The machinery is the sublattice-resolved finite-difference treatment: exchange is split into intra-sublattice ferromagnetic exchange, inhomogeneous inter-sublattice exchange, and a homogeneous antiferromagnetic exchange proportional to $4A_0/(M_S a^2)$ acting on the other sublattice's magnetization, with Neumann boundary conditions, Eq. (5), that couple the two sublattices and the DMI at surfaces. Magnetoelasticity is added by solving the elastodynamic equation $\\rho\\,\\partial_t^2\\mathbf{u}=\\mathbf{f}_{\\mathrm{tot}}$ with the elastic force computed as the numerical divergence of stress, and by coupling each sublattice magnetization to the shared strain through its own magnetoelastic constants $B_1^{(s)}$ and $B_2^{(s)}$. Time integration uses embedded Runge-Kutta methods with adaptive stepping.","core_discovery":"The central claim is that a finite-difference micromagnetics code that evolves each sublattice magnetization with the Landau-Lifshitz-Gilbert torque, spin-transfer torques, antiferromagnetic exchange, Dzyaloshinskii-Moriya interaction, and magnetostatic fields, while simultaneously solving the elastodynamic equation with magnetoelastic forces, produces the three demonstrated behaviors: grain-boundary-dependent domain-wall mobility consistent with a known analytical model, stepwise domain-wall transport in NiO under an 8 GHz shear traction pulse, and Mn3Sn stray-field images comparable to measured nitrogen-vacancy magnetometry scans.","pith_inferences":["The same finite-difference machinery should extend to other non-collinear magnets, such as the wider Mn3X family, and to coupled elastic-magnetic wave devices like surface-acoustic-wave-driven domain-wall motion; the paper does not test those cases.","A direct sensitivity test would be to vary the exchange and DMI parameters used for the Mn3Sn simulation and check whether the patchwork stray-field pattern persists; if it changes drastically, the agreement with the measured image is parameter-fitted rather than predictive.","The magnetoelastic formulation with a shared elastic displacement but per-sublattice coupling constants could be used to study strain-induced spin reorientation transitions or magnetoacoustic resonance in antiferromagnets, phenomena not demonstrated in the paper.","The comparison with the nitrogen-vacancy images is qualitative; quantifying local stray-field statistics point by point would provide a stronger, more quantitative test of the model."],"forward_implications":["The same solver can model collinear antiferromagnets, non-collinear antiferromagnets, and ferrimagnets, with each sublattice having its own anisotropy, DMI tensor, and magnetoelastic constants while sharing one elastic displacement field.","A strain pulse generated by boundary traction can move domain walls stepwise between notches, demonstrating the working principle of a traction-driven antiferromagnetic racetrack memory.","The simulated out-of-plane stray field 60 nm above a polycrystalline Mn3Sn film has the same patchwork character and order of magnitude as nitrogen-vacancy magnetometry scans.","Current-driven domain-wall velocity in a polycrystalline antiferromagnet decreases and pinning strengthens as the inter-grain exchange is reduced, with the zero-reduction case matching the analytical model.","Because the code is open-source and object-oriented, new physical terms can be added by users without rewriting the core solver."],"supporting_citations":[{"why":"Supplies the antiferromagnetic exchange field expression used in Eq. (3) and the analytical domain-wall mobility model matched by the zero-grain simulation.","marker":"[31]"},{"why":"Provides the Voronoi grain-boundary reduction method and the polycrystalline permalloy results that the antiferromagnetic mobility curves are compared with.","marker":"[30]"},{"why":"Defines the base solver design and verification benchmarks that the new implementation is tested against.","marker":"[5]"},{"why":"Provides the experimental nitrogen-vacancy magnetometry image of Mn3Sn that the simulated stray-field map is compared with.","marker":"[47]"},{"why":"Supplies the prior finite-difference magnetoelastic simulator that the elastodynamic and magnetoelastic implementation is adapted from.","marker":"[66]"},{"why":"Gives the scheme for antiferromagnetic magnetoelastic coupling with per-sublattice magnetoelastic constants.","marker":"[73]"},{"why":"Provides the DMI energy and boundary-condition formalism used for the Neumann conditions in Eq. (5).","marker":"[60]"},{"why":"Supplies the magnetoelastic and stiffness parameters for the NiO racetrack simulation.","marker":"[41]"}],"fun_headline_variants":["Mumax+ simulates antiferromagnetic racetrack memory on GPU","Antiferromagnets, meet GPU micromagnetics: it's mumax+","Extensible GPU code mumax+ models complex magnetic orderings","Mumax+ reproduces antiferromagnetic racetrack dynamics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the finite-difference exchange field and Neumann boundary conditions of the antiferromagnetic model faithfully represent real antiferromagnets on a grid; in the Mn3Sn demonstration this premise includes material parameters that are presented without a cited source.","fun_headline_variants_meta":{"raw":{"variants":["Mumax+ simulates antiferromagnetic racetrack memory on GPU","Antiferromagnets, meet GPU micromagnetics: it's mumax+","Extensible GPU code mumax+ models complex magnetic orderings","Mumax+ reproduces antiferromagnetic racetrack dynamics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001654,"raw_usage":{"total_tokens":6505,"prompt_tokens":820,"completion_tokens":5685,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":436,"completion_tokens_details":{"reasoning_tokens":5614}},"tokens_in":436,"tokens_out":5685,"duration_ms":36096,"temperature":1.0,"reasoning_tokens":5614,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:26:03.227331+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Send a single calibrated 8 GHz shear traction pulse through a NiO strip with notches at 300 nm and 600 nm: the model predicts each domain wall moves exactly one notch spacing, 150 nm, so if a single pulse moves the walls by zero or by more than one notch, the coupled magnetoelastic implementation is wrong.","supporting_citations":[{"cited_title":"Journal of Applied Physics 115(23), 233903 (2014) https://doi.org/10.1063/1.4883297 https://pubs.aip.org/aip/jap/article- pdf/doi/10.1063/1.4883297/15138111/233903 1 online.pdf","cited_arxiv_id":null,"evidence_quote":"Provides the Voronoi grain-boundary reduction method and the polycrystalline permalloy results that the antiferromagnetic mobility curves are compared with."},{"cited_title":"Accessed 2025-06-10","cited_arxiv_id":null,"evidence_quote":"Provides the experimental nitrogen-vacancy magnetometry image of Mn3Sn that the simulated stray-field map is compared with."},{"cited_title":"Open Research Europe 1(35) (2021) https://doi.org/10","cited_arxiv_id":null,"evidence_quote":"Supplies the prior finite-difference magnetoelastic simulator that the elastodynamic and magnetoelastic implementation is adapted from."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the scheme for antiferromagnetic magnetoelastic coupling with per-sublattice magnetoelastic constants."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the DMI energy and boundary-condition formalism used for the Neumann conditions in Eq. (5)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the magnetoelastic and stiffness parameters for the NiO racetrack simulation."}],"review_version":1}