Combining Quasiparticle Self-Consistent G ​ W GW and Machine-Learned DFT+ U U in Search of Half-Metallic Heuslers Preprint: APS/123-QED Zefeng Cai, Malcolm J. A. Jardine, Maituo Yu, Chenbo Min, Jiatian Wu, Hantian Liu, Derek Dardzinski, Christopher J. Palmstrøm, Noa Marom Abstract Half-metallic Heusler compounds are of significant interest for spintronics. For device fabrication, compounds that can be epitaxially grown on III-V semiconductors are particularly attractive. We present a first-principles investigation of four Co-based and two Ni-based Heusler compounds that are lattice-matched to InAs. The results of density functional theory (DFT) using semi-local and hybrid functionals are compared to quasiparticle self-consistent G ​ W GW (QPGW). We also consider DFT with machine-learned Hubbard U U corrections [npj Computational Materials 6, 180 (2020)] with a new Bayesian optimization (BO) objective function to determine the U U values that yield the closest agreement with the QPGW band structure and magnetic moments. We find that DFT+U(BO) can adequately reproduce the key QPGW features in most cases. Our results reveal a strong method dependence of the degree of spin polarization at the Fermi level and, in some cases, even the dominant spin channel (majority or minority). Of the materials studied here, Co 2 TiSn and Co 2 ZrAl are the most likely to be half-metals, and Co 2 MnIn is likely to be a near-half-metal. I Introduction The Heuslers are a large family of ternary compounds with the formula X 2 YZ, where X and Y are transition metals or rare earth elements and Z is a p -block element Knowlton and Clifford 1912 . The sustained interest in Heuslers has been driven by their useful properties for a wide array of applications Felser and Hirohata 2015 ; Graf et al. 2011 ; Felser et al. 2015 ; Farshchi and Ramsteiner 2013 ; Wollmann et al. 2017 ; Graf et al. 2010 ; Hirohata and Lloyd 2022 . Magnetic Heusler compounds can serve as a ferromagnetic contact in magnetic tunnel junctions and magnetic spin valves Peterson et al. 2016 ; Sahoo et al. 2016 . This can significantly improve the tunneling magnetoresistance (TMR) Kämmerer et al. 2004 ; Okamura et al. 2005 , which is a key metric for efficient magnetic transport in a hard drive or a magnetic random access memory (MRAM) chip Elphick et al. 2021 ; Palmstrøm 2016 . Furthermore, the high spin polarization displayed by some Heuslers would make them ideal spin contacts to the semiconductor in spin-field effect transistors (FETs) Datta and Das 1990 , spin light emitting diodes Holub et al. 2007 , and spin-lasers Bhattacharya et al. 2010 . More recently, some members of the Heusler family have been found to be topological insulators Yan and Felser 2017 ; Chadov et al. 2010 . Here, we focus on Heusler compounds that may exhibit half-metallicity De Groot et al. 1983 ; Ishida et al. 1995 ; Galanakis et al. 2008 ; Felser and Hirohata 2015 ; Nishihaya et al. 2025 , meaning that one spin channel is metallic and the other is semiconducting. Half-metals are of interest for spintronics thanks to the promise of achieving close to 100% spin polarization at the Fermi level. Chemically ordered films and atomically sharp, defect-free interfaces are desirable for spintronic devices because the preservation of half-metallicity in Heusler alloys and spin-dependent transport across interfaces are highly sensitive to disorder, interfacial roughness, and impurities Elphick et al. 2021 ; Palmstrøm 2016 ; Orgassa et al. 1999 ; Van Roy et al. 2000 ; Sakuraba et al. 2005a ; Adelmann et al. 2005 ; Marukame et al. 2007 ; Schultz et al. 2009 ; Guillemard et al. 2020 ; De Boeck et al. 2002 ; Parkin et al. 2004 ; Yuasa et al. 2004 ; Velev et al. 2008 ; Wang et al. 2010 ; Hirohata et al. 2022 . In particular, deviations from ideal L2 1 ordering in full-Heusler compounds have been shown to strongly suppress spin polarization Elphick et al. 2021 and increase magnetic damping Guillemard et al. 2020 . Interfacial defects and disorder may introduce spin-flip scattering that reduces tunneling magnetoresistance and spin injection efficiency De Boeck et al. 2002 ; Wang et al. 2010 ; Hirohata et al. 2022 . It has also been reported that half-metallicity may be degraded near surfaces and interfaces Picozzi et al. 2003 ; Khosravizadeh et al. 2009 ; Galanakis and Mavropoulos 2007 ; Galanakis et al. 2005 ; Galanakis 2002 ; Jenkins 2004 ; Jenkins and King 2002 . It has been found that the properties of the interface are governed by the local magnetic ordering and electronic structure of both the Heusler compound and the semiconductor Palmstrøm 2016 ; Rath et al. 2018 ; Heischmidt et al. 2023 . These considerations motivate the search for Heuslers that can form epitaxial interfaces with appropriate substrates via growth techniques that enable precise stoichiometric and structural control. Ordered Heusler films can be fabricated by high temperature annealing when in contact with thermodynamically stable metals or insulating oxides Sakuraba et al. 2005b ; Tezuka et al. 2006 . However, growing Heuslers on reactive semiconductors requires the use of molecular-beam epitaxy (MBE) Hirohata et al. 2006 with individual elemental sources. Heusler compounds have been grown on various III-V semiconductors with a compatible FCC lattice. Thanks to the large compositional space of the Heusler family, which may be further expanded by substitutional alloying, it is possible to find lattice matched Heuslers for substrates with a wide range of lattice parameters including GaAs, InP, InAs, GaSb, and InSb Palmstrøm 2016 ; Alijani et al. 2011 ; Gao et al. 2013 ; Valerio et al. 2005 . The growth of epitaxial Heusler/III-V interfaces is nontrivial because the elements comprising the Heusler (such as Mn) may diffuse into and react with the III-V semiconductor at the interface Law et al. 2024 ; Nishihaya et al. 2025 ; Singh et al. 2006 , or conversely, elements from the semiconductor may out-diffuse into the Heusler film Hamaya and Yamada 2022 , requiring the use of barrier layers. Here, we focus on Heusler compounds predicted to have good lattice matching with InAs. InAs is attractive for spintronics and quantum computing applications due to its high electron mobility, large effective g -factor, strong spin-orbit coupling, and the absence of a Schottky barrier at the interface with various metals Zhang et al. 2023 ; Prabhakar et al. 2011 ; Takase et al. 2017 ; Feng et al. 2016 . Because GaSb has a close lattice parameter to InAs, the materials considered here are also matched to GaSb. Considering the high cost of MBE, computer simulations can help select the most promising candidate materials for growth experiments. There is a long history of first principles studies of Heusers and related half-Heuslers starting from the seminal paper by de Groot et al. , which established the concept of half-metallicity in half-Heusler compounds De Groot et al. 1983 . Subsequent experimental work confirmed half-metallic behavior in materials such as NiMnSb Hanssen et al. 1990 ; Hanssen and Mijnarends 1986 ; Kirillova et al. 1995 . This discovery motivated extensive computational investigations, predominantly using density functional theory (DFT) with local or semi-local exchange–correlation functionals Ishida et al. 1995 ; Galanakis et al. 2002a ; Galanakis et al. 2002b . In recent years, high-throughput computational screening has been employed to discover and design new Heusler candidates for diverse applications Roy et al. 2012 ; Yan and de Visser 2014 ; Sanvito et al. 2017 ; Rotjanapittayakul et al. 2018 ; Ma et al. 2017 ; Ma et al. 2018 ; Jiang and Yang 2021 ; Jiang et al. 2022 ; Jiang and Yang 2023 ; Hilgers et al. 2025 . One limitation of such studies is that the predicted properties of Heusler compounds (and related classes of materials), including band gaps and magnetic properties, are highly sensitive to the choice of approximation for the exchange–correlation functional. Local and semi-local functionals may fail to reliably describe d -electron magnetism and half-metallicity in Heuslers and related materials Nawa and Miura 2019 ; Kandpal et al. 2006 ; Faleev et al. 2017a ; Faleev et al. 2017b ; Faleev et al. 2017c ; Anisimov et al. 1997 ; Kobayashi et al. 2008 ; Heischmidt et al. 2023 . The deficiencies of (semi-)local DFT functionals may be attributed to a combination of the self-interaction error (SIE) and an inadequate treatment of strong correlation effects, both particularly pronounced for localized d - and f -electrons Perdew and Zunger 1981 ; Mori-Sánchez et al. 2008 ; Cohen et al. 2008 ; Kotliar et al. 2006 . Hybrid functionals, in which a fraction of exact (Fock) exchange is mixed with semi-local exchange and correlation, often produce more accurate band gaps and band structures Shi et al. 2017 ; Bilc and Ghosez 2011 . However, hybrid functionals may overestimate exchange splitting and magnetic moments in transition-metals Jana et al. 2018 ; Paier et al. 2006 ; Fu and Singh 2019 ; Janthon et al. 2014 ; Zahedifar and Kratzer 2018 , which are the most common components in Heusler compounds, and may inaccurately describe spectral features near the Fermi level Gao et al. 2016 ; Coulter et al. 2013 . Further improvement may be achieved by going beyond DFT. Many-body perturbation theory in the GW approximation Golze et al. 2019 , where G represents the Green’s function and W the screened Coulomb interaction, provides a more reliable description of correlated and magnetic materials Acharya et al. 2021 ; Jiang 2015 ; Coulter et al. 2013 ; Yamasaki and Fujiwara 2003 ; Kotani and van Schilfgaarde 2008 , including Heusler and half-Heusler compounds Zahedifar and Kratzer 2018 ; Gürbüz et al. 2023 ; Tas et al. 2017 ; Nishihaya et al. 2025 . In particular, the quasiparticle self-consistent GW (QPGW) method Faleev et al. 2004 ; van Schilfgaarde et al. 2006 ; Shishkin et al. 2007 eliminates the dependence on the mean-field starting point, which is a known limitation of one-shot G_{0}W_{0} Cunningham et al. 2023 ; Kotani and Kino 2009 ; Kotani et al. 2007 ; Friedrich et al. 2022 ; Ismail-Beigi 2017 . However, the high computational cost of hybrid functionals, and even more so of GW , limits the applicability of these methods to relatively small systems and precludes their utilization for studies of large interface models. The DFT+U method provides a balance between accuracy and efficiency by adding Hubbard U corrections that introduce an orbital-dependent on-site Coulomb interaction to (semi-)local functionals Dudarev et al. 1998 . DFT+U with an appropriate choice of the system-dependent U parameter(s) has been widely applied to correlated materials with localized d or f orbitals, such as transition-metal compounds Himmetoglu et al. 2014 ; Zhou et al. 2004 ; Aryasetiawan et al. 2006 ; Mosey et al. 2008 ; Kulik et al. 2006 ; Jiang et al. 2010 ; Kick et al. 2019 . For Heusler compounds and related materials, introducing Hubbard U corrections has been shown to improve the resulting electronic structure, including magnetic moments, spin polarization and band widths Kandpal et al. 2006 ; Kandpal et al. 2007 ; Wurmehl et al. 2005 ; Khosravizadeh et al. 2009 ; Nourmohammadi and Abolhasani 2010 ; Şaşıoğlu et al. 2013 ; Fecher and Felser 2007 ; Mellah and Demmouche 2022 ; Baral et al. 2015 ; Sasıoglu et al. 2025 . We have developed a method of machine learning the U values for a given material using Bayesian optimization (BO) with an objective function that aims to reproduce as closely as possible the band structure obtained from a higher-level reference method such as a hybrid functional or QPGW Yu et al. 2020 . The DFT+U(BO) method has enabled us to perform large-scale simulations of surfaces and interfaces Yang et al. 2021 ; Yu et al. 2021 ; Yang et al. 2022 ; Jardine et al. 2023 ; Jardine et al. 2025 , including Heusler/ III-V interfaces Heischmidt et al. 2023 , which would not have been possible otherwise. Here, we investigate the electronic and magnetic properties of four Co-based and two Ni-based Heusler compounds, whose lattice parameters are matched to InAs (or GaSb), as summarized in Table 1 . We compare the results obtained from QPGW to DFT with the generalized gradient approximation (GGA) of Perdew, Burke, and Ernzerhof (PBE) Perdew et al. 1996 ; Perdew et al. 1997 and the range-separated hybrid functional of Heyd-Scuseria-Ernzerhof (HSE) Heyd et al. 2003 ; Heyd et al. 2006 , as well as PBE+U(BO). For DFT+U(BO), we have formulated a new objective function that includes the atomic magnetic moments in addition to the band structure, and we use QPGW as the reference method. The optimal U values obtained from the BO procedure enable PBE+U(BO) to reproduce the key qualitative features of the QPGW electronic structure. We find that the choice of electronic structure method strongly influences the results. In some cases, different methods even provide qualitatively different predictions for whether a certain material is a half-metal and whether the spin polarization at the Fermi level is up or down. Of the materials studied here, only Co 2 TiSn is unanimously predicted by all methods to be a half metal. Co 2 ZrAl is predicted by all methods except HSE to be a half metal. Co 2 MnIn is predicted by QPGW to be a near-half-metal with a high degree of minority spin polarization at the Fermi level. Our findings have significant implications for the reliability of high-throughput studies of magnetic Heuslers based on semi-local DFT. Table 1: Summary of the computed equilibrium lattice constant, total magnetic moment, spin polarization at the Fermi level, and effective U values applied to the d orbitals of the X and Y elements of each Heusler compound (a U correction was not applied to the Z element). The results of different electronic structure methods are compared for the total magnetic moment and the spin polarization at the Fermi level. A negative spin polarization indicates that the minority spin channel is dominant at the Fermi level. Material a_{0} (Å) Total Magnetic Moment ( \mu_{B} ) Spin Polarization (%) U (eV) PBE HSE QPGW PBE+U PBE HSE QPGW PBE+U X Y Co 2 MnIn 5.980 4.455 7.374 6.448 5.853 85.98 -64.36 -94.62 -94.29 2.252 0.751 Co 2 MnSn 5.984 5.030 6.295 5.853 5.257 98.68 -82.28 -66.67 94.71 1.512 1.652 Co 2 TiSn 6.082 1.997 2.009 1.996 1.997 100.00 100.00 100.00 99.98 3.534 3.654 Co 2 ZrAl 6.063 0.999 1.304 1.000 0.999 100.00 -32.83 100.00 100.00 1.331 9.379 Ni 2 MnSb 6.059 4.044 4.462 4.662 4.620 -30.77 14.94 -43.03 7.85 -1.512 5.155 Ni 2 MnSn 6.056 4.108 4.613 4.780 4.466 -67.52 -65.63 -69.93 -56.95 2.392 2.352 II Methods II.1 DFT+U(BO) Previously, we had proposed Bayesian optimization as a practical approach for finding the optimal Hubbard U parameters for a given material within the Dudarev formulation of DFT+U Dudarev et al. 1998 . The U corrections can be applied to any (semi-)local functional. Here, we use PBE. The BO objective function is formulated to reproduce as closely as possible the results of a reference method Yu et al. 2020 . Originally, the objective function included the first two terms of Eq. 1 , i.e., the difference in the band gap, \Delta Gap (only used for semiconductors), and the difference in the band structure, \Delta Band, weighted by \alpha_{1} and \alpha_{2} , respectively Yu et al. 2020 . For magnetic materials, we now introduce a third term, \Delta Mag, representing the difference in atomic magnetic moments, with an additional weight \alpha_{3} : \displaystyle{f(\vec{U})}= \displaystyle-{\alpha_{1}}(\Delta\text{Gap})^{2} (1) \displaystyle-{\alpha_{2}}(\Delta\text{Band})^{2} \displaystyle-{\alpha_{3}}(\Delta\text{Mag})^{2} where \vec{U} = [ U^{1} , U^{2} , …, U^{n} ] represents the vector of U values applied to different elements and U^{i}\in[-10,10] eV. \Delta\text{Gap} is defined as: \Delta\text{Gap}=E_{\text{g}}^{\text{ref}}-E_{\text{g}}^{\text{PBE+U}} (2) \Delta\text{Band} is defined as an extension of the previous definition of the root mean squared difference between the PBE+U and the reference band structure Yu et al. 2020 ; Huhn and Blum 2017 : \Delta\text{Band}=\sqrt{\begin{aligned} \frac{1}{N_{E}}\sum^{N_{k}}_{i=1}\sum^{N_{b}}_{j=1}w_{i}\Bigl(&(\epsilon_{\text{ref}}^{j}[k_{i}]-\epsilon_{\text{ref},0}^{j})\\ -&(\epsilon_{\text{PBE+U}}^{j}[k_{i}]-\epsilon_{\text{PBE+U},0}^{j})\Bigr)^{2}\end{aligned}} (3) Here, N_{E} is the total number of eigenvalues included in the comparison, N_{k} is the number of k -points, and N_{b} is the number of bands selected for comparison. The weight factor w_{i} reflects the relative importance of each k -point based on its contribution to the Brillouin zone (BZ) volume. For k -paths with an equal number of k -points, it is normalized to ensure uniform sampling across the BZ: \forall\,i\in I,\quad w_{i}=\frac{|\mathbf{k}_{I}|}{\sum_{J}|\mathbf{k}_{J}|}, (4) I and J are the index sets of k -points along the k -paths \mathbf{k}_{I} and \mathbf{k}_{J} , respectively. |\mathbf{k}_{I}| denotes the length of the k -path \mathbf{k}_{I} in reciprocal space. Here, \epsilon_{\text{ref},0}^{j} and \epsilon_{\text{PBE+U},0}^{j} denote reference energy levels defined for band j in the reference method and in the PBE+U calculation, respectively. These reference energies are introduced to account for the fact that absolute energy levels are not directly comparable between calculations performed with different exchange-correlation functionals or pseudopotentials; only relative energy differences are physically meaningful. The reference level can be chosen by the user. common choices are the respective vacuum level or the valence band maximum (VBM) in both methods. For metallic systems, the reference energy is typically taken to be the Fermi level of each calculation. In the same vein, \Delta Mag is defined as the root mean squared error of the PBE+U magnetic moments compared to the reference method: \Delta\text{Mag}=\sqrt{\frac{1}{N_{i}}\sum^{N_{i}}_{i=1}(m_{\text{ref}}^{i}-m_{\text{PBE+U}}^{i})^{2}} (5) where N_{i} is the total number of atoms, and m is the magnetic moment produced by each method. Including the magnetic term leads to faster and more robust convergence of the BO algorithm, as shown in the SI. Previously, we used HSE as the reference method Yu et al. 2020 . However, as discussed above, for the materials studied here, we consider QPGW as a more reliable choice. For metallic Heusler compounds, the objective function coefficients are set to \alpha_{1}=0.0 , \alpha_{2}=0.5 , and \alpha_{3}=0.5 , assigning equal weights to the band structure and magnetic moments. N_{b} is set to 16 with 8 bands above and 8 bands below the Fermi level. The parameter \kappa , which controls the balance between exploration and exploitation in the BO upper confidence bound acquisition function Yu et al. 2020 , is set to 5. For the Heuslers studied here, Hubbard U corrections were applied to the d -orbitals of the transition-metal elements, X and Y, but not to the main-group element, Z. The BO algorithm was deemed converged if either (i) the difference in the objective function between two consecutive iterations was less than 0.0001 or (ii) the change in the interpolated optimum of the objective function over ten successive iterations was less than 0.005. The resulting U values for each material are summarized in Table 1 and BO convergence plots are provided in the SI. II.2 Computational Details All calculations were performed using the Vienna Ab initio Simulation Package (VASP) with the projector augmented wave (PAW) method Kresse and Hafner 1993 ; Kresse and Joubert 1999 ; Blöchl 1994 . All calculations used a plane-wave basis set with a kinetic energy cutoff of 400 eV and an energy convergence criterion of 10^{-6} eV. Spin-orbit coupling (SOC) was used for all calculations Mavropoulos et al. 2004 , including QPGW (an analysis of the effect of SOC on the electronic structure is provided in the SI). The recommended pseudopotentials provided by VASP (standard and GW ) were used for all calculations. We included non-spherical contributions from the density gradient in the PAW spheres and applied a cutoff to the gradient field to restore the full lattice symmetry in order to account for magnetic anisotropy. The Brillouin zone was sampled using the \Gamma -centered scheme. For QPGW calculations, a 7\times 7\times 7 k -point mesh was used. For PBE and HSE calculations a k-point mesh of 9\times 9\times 9 was used. Structural relaxation was performed with PBE using the conjugated gradients algorithm. The ionic relaxation was terminated when the change in total energy between two successive ionic steps fell below 10^{-5} eV. The resulting equilibrium lattice constants of all Heusler compounds studied here are listed in Table 1 . The QPGW Shishkin et al. 2007 ; Shishkin and Kresse 2007 calculations were performed self-consistently with HSE used as the starting point. Based on convergence tests provided in the SI, seven self-consistent GW steps were performed. The total number of bands was set to 94 or 96, depending on the parallelization setup, corresponding to approximately twice the number of occupied bands for all six materials. The number of bands for which the self-energy shift was calculated was set to 78. The number of imaginary frequency and imaginary time grid points was set to 140. QPGW band structures were calculated via Wannier interpolation using the Wannier90 code Pizzi et al. 2020 . The on-site magnetic moments were obtained by decomposing the Kohn-Sham orbitals onto the PAW projectors. The spin polarization percentage at the Fermi level was calculated as \mathrm{SP}\%=100\,P(E_{f}) , where P(E_{f}) is given by: {P(E_{f})}=\frac{\int_{E_{f}-\Delta}^{E_{f}+\Delta}\left[D_{\uparrow}(E)-D_{\downarrow}(E)\right]\mathrm{d}E}{\int_{E_{f}-\Delta}^{E_{f}+\Delta}\left[D_{\uparrow}(E)+D_{\downarrow}(E)\right]\mathrm{d}E} (6) Here, D_{\uparrow} and D_{\downarrow} are the spin-up (majority) and spin-down (minority) density of states (DOS). The integration is performed over an energy window of 2\Delta=3.5\,kT=0.091\,\mathrm{eV} centered at the Fermi level, chosen to match the full width at half maximum (FWHM) of the derivative of the Fermi-Dirac distribution at room temperature, thereby capturing the thermal broadening of the Fermi edge Wegner et al. 2006 . A sensitivity analysis of the SP value to the choice of integration window is provided in the SI. A negative sign of SP% indicates a minority spin polarization at the Fermi level. For the PBE, HSE, and PBE+U calculations, element- and orbital-decomposed band structures were obtained by projecting the Kohn–Sham states onto PAW on-site partial-wave projectors. Band structure and DOS plots were generated using the open-source package vaspvis Dardzinski et al. 2022 , which is available on the Python Package Index (PyPI) via pip install vaspvis , and on GitHub at https://github.com/caizefeng/vaspvis . The DFT+U(BO) calculations were performed using the updated BayesOpt4dftu package, accessible at https://github.com/caizefeng/BayesianOpt4dftu . III Results and discussion Figure 1: Band structure and element-resolved DOS of Co 2 MnSn calculated using (a) PBE, (b) HSE, (c) QPGW, and (d) PBE+U(BO). In the band-structure panels, the majority and minority spin channels are shown in blue and red, respectively. In the DOS panels, the majority and minority spin components are plotted on the left and right sides, respectively. The rightmost panels present a magnified view of the DOS around the Fermi level. The total magnetic moment (MM, in \mu B ) and spin polarization percentage at the Fermi level (SP%) are listed for each method. Negative spin polarization indicates that the minority spin channel is dominant around the Fermi level. The dashed box in (c) highlights the flat minority-spin band near the Fermi level that gives rise to the minority-dominated spin polarization in the QPGW results. The Heusler alloys selected in this study are Co 2 MnIn, Co 2 MnSn, Co 2 TiSn, Co 2 ZrAl, Ni 2 MnSb, and Ni 2 MnSn. These compounds have a lattice mismatch below 1% with InAs or GaSb. Further screening criteria were applied to ensure that the compounds selected for this study possess a stable ferromagnetic ground state with the Fm \bar{3} m symmetry at room temperature. It is worth noting that several other Co-, Ni-, or Ru-based full Heusler alloys also have nearly perfect lattice matching with InAs, but do not meet the magnetic stability criteria. For example, Co 2 CrIn has a mismatch of only 0.02% with InAs but has been reported to exhibit a ferrimagnetic ground state Wurmehl et al. 2006 . Co 2 NbSn, with a 0.89% mismatch to GaSb, has a low Curie temperature of 116 K and favors a competing Pmma structure Wolter et al. 2002 . Ru 2 MnAl, Ru 2 MnGa, and Ru 2 MnGe have excellent lattice matching to InAs, but prefer an antiferromagnetic state at room temperature Gotoh et al. 1995 . Table 1 presents a summary of the total magnetic moments and the degree of spin polarization around the Fermi level obtained with different electronic structure methods. According to all methods considered here, the Ni-based compounds do not have a gap in either spin channel and are not strongly spin-polarized around the Fermi level. Therefore, we focus the discussion in the main text on the Co-based compounds and provide the complete results for the Ni-based compounds in the SI. Co 2 MnSn and Co 2 MnIn exhibit relatively large magnetic moments due to the presence of the magnetic Mn atom in addition to Co. For these materials, the magnetic moments produced by different electronic structure methods vary significantly, with PBE giving the lowest value, HSE giving the highest value, and QPGW in between. For Co 2 MnSn, experimental studies report total magnetic moments in the range of 5.05-5.12 \mu_{B} Webster 1971 ; Brown et al. 2000 ; Baral et al. 2015 , which are closest to the PBE result of 5.03 \mu_{B} and the PBE+U(BO) value of 5.26 \mu_{B} . Co 2 TiSn and Co 2 ZrAl have smaller magnetic moments because the Y-site element is non-magnetic. For these materials the results of different electronic structure methods are more consistent. For Co 2 TiSn, all methods produce a total magnetic moment of \sim 2 \mu_{B} , in excellent agreement with experimental measurements Chandra Kandpal et al. 2007 ; Bainsla and Suresh 2016 ; Barth et al. 2010 ; Ooka et al. 2017 ; Webster and Ziebeck 1973 . For Co 2 ZrAl, all methods (except HSE) predict a magnetic moment close to 1 \mu_{B} , which is slightly higher than the experimentally measured value of approximately 0.7 \mu_{B} Kanomata et al. 2005 ; Buschow and van Engen 1981 ; Ziebeck and Webster 1974 . The measured magnetic moments of Ni 2 MnSb range from 3.22 to 4.31 \mu_{B} Shinohara 1970 ; Buschow et al. 1983 . The scatter may be due to differences in sample quality or stoichiometry. The computed values range from 4.04 \mu_{B} with PBE to 4.66 \mu_{B} with QPGW. Overall, both HSE and QPGW tend to overestimate the total magnetic moments relative to experiment for several compounds. Of the four Co-based compounds, only Co 2 TiSn is unanimously predicted by all methods to be a half-metal. In the following, we present a detailed analysis of the origin of the differences between electronic structure methods. Figure 1 shows the band structure and element-resolved DOS of Co 2 MnSn obtained with PBE, HSE, QPGW, and PBE+U(BO). With the PBE functional, the DOS around the Fermi level is mainly contributed by the Co 3d and Mn 3d states. The DOS shows a half-metallic gap of 0.406 eV in the minority-spin channel (on the right-hand side of the DOS panels) and located 0.162 eV above the Fermi level. It should be noted that the position and width of the gap are determined from the band structure rather than from the DOS. Although the minority-spin gap appears sizable and may seem to intersect the Fermi level in the DOS plots of Figure 1 (a), as well as in Figure 3 (a) and Figure 5 (a), a small but nonzero minority-spin density remains at the Fermi level, leading to a spin polarization below 100% (98.7% in this case). The apparent discrepancy arises because the band structure is plotted along selected high-symmetry paths, which do not necessarily represent the full Brillouin-zone integration that determines the total DOS. The HSE functional yields a considerably different band structure than PBE, as shown in panel b. The Mn d states are strongly exchange-split and shifted farther in energy above and below the Fermi level. The gap of 2.40 eV in the minority spin channel is considerably wider than with PBE and it is located 0.509 eV above the Fermi level. The DOS around the Fermi level is contributed predominantly by the Co 3d states, leading to minority spin polarization of -82.3%. The behavior of HSE likely originates from the overestimation of exchange splitting in d -electron systems by the non-local exchange term, which is a known limitation of this functional Paier et al. 2006 ; Paier et al. 2007 . With QPGW (panel c), the exchange splitting of the Mn d states is not as large as with HSE. The gap in the minority spin channel is 0.818 eV and it is located 4.756 eV above the Fermi level. The DOS around the Fermi level is dominated by the Co d states, similar to HSE, with a small contribution from the Mn d states. This results in a minority spin polarization of -66.7%. Although the QPGW band structure appears overall more similar to PBE than to HSE, a crucial difference is the rather flat minority band that lies slightly above the Fermi level around the X-point (indicated by the box in panel c). This is sufficient to flip the spin polarization around the Fermi level. The results of PBE+U(BO) are in between PBE and QPGW, as shown in panel d. The gap in the minority spin channel is 0.905 eV, similar to QPGW. The DOS around the Fermi level is dominated by the Co d manifold. However, the flat minority band around the X-point lies below the Fermi level, leading to a majority spin polarization of 94.7%, similar to PBE. The interplay between the relative positions of the Co and Mn states d can thus lead to qualitatively different results with different electronic structure methods. A spin polarization of about 60% has been measured for Co 2 MnSn by point-contact Andreev reflection (PCAR) Varaprasad et al. 2009 . We note that PCAR measures only the absolute value of the spin polarization irrespective of its sign Soulen et al. 1998 ; Nadgorny et al. 2003 . The absolute value of the spin polarization obtained from QPGW is close to the PCAR experiment, but we cannot determine whether the minority spin polarity is correct. This leaves us with the questions of whether Co 2 MnSn is a half-metal, what is the dominant spin channel at the Fermi level, and which method should we trust? Figure 2: Orbital-resolved DOS of Co 2 MnSn calculated using (a) PBE, (b) HSE, (c) QPGW, and (d) PBE+U(BO), compared to the resonant photoemission spectroscopy (RPES) spectra of Co-3 d and Mn-3 d , reproduced with permission from Ref. Baral et al. 2015 . The computed spectra are broadened by a Gaussian with \sigma=0.77\,\mathrm{eV} to simulate the resolution of the experiment. All spectra are normalized to the same maximum intensity. The satellite features between -6\,\mathrm{eV} and -2\,\mathrm{eV} in the experimental spectrum of Co-3 d are attributed in Ref. Baral et al. 2015 to excitation effects that are not captured by the electronic structure methods used here. To attempt to answer these questions, the orbital-resolved DOS of Co 2 MnSn is compared with resonant photoemission spectroscopy (RPES) data Baral et al. 2015 in Figure 2 . The experimental data shown are difference spectra that were obtained therein by subtracting the off-resonance spectrum from the on-resonance spectrum to resolve the partial DOS of the Co-3 d and Mn-3 d states. The RPES data shows that the DOS close to the Fermi level is predominantly contributed by the Co-3 d , whereas the Mn-3 d states lie lower in energy, around -3.8 eV below the Fermi level. The satellite feature between -6 eV and -2 eV in the Co-3 d RPES spectrum was attributed in Ref. Baral et al. 2015 to excited-state effects that are not captured by the electronic structure methods used here. We focus the comparison to experiment on the Co-3 d and Mn-3 d main peak positions. With PBE the Co-3 d and Mn-3 d overlap, which clearly contradicts the experimental data. HSE shifts the Mn-3 d peak to around -4 eV, which is close to the experiment, but the Co-3 d states are shifted too deep to around -2.6 eV below the Fermi level, which deviates significantly from the experiment. With QPGW the Mn-3 d peak is slightly too high and the Co-3 d peak is slightly too low, providing the closest match to experiment of the methods considered here. The PBE+U(BO) partial DOS is very similar to QPGW. In Ref. Baral et al. 2015 it was found that applying a Hubbard U correction of 3 eV to the Mn-3 d states and no correction to the Co-3 d provided the closest agreement with experiment. This is different than the U values of 1.512 eV for Co-3 d and 1.652 eV for Mn-3 d , which best reproduce the QPGW results. Detailed experimental characterization with spin-resolved angle-resolved photoemission spectroscopy (ARPES) Zhang et al. 2022 ; Iwasawa et al. 2024 would be needed to unambiguously determine which electronic structure method is the most reliable. Figure 3: Band structure and element-resolved DOS of Co 2 MnIn calculated using (a) PBE, (b) HSE, (c) QPGW, and (d) PBE+U(BO). In the band-structure panels, the majority and minority spin channels are shown in blue and red, respectively. In the DOS panels, the majority and minority spin components are plotted on the left and right sides, respectively. The rightmost panels present a magnified view of the DOS around the Fermi level. The total magnetic moment (MM, in \mu B ) and spin polarization percentage at the Fermi level (SP%) are listed for each method. Negative spin polarization indicates that the minority spin channel is dominant around the Fermi level. The case of Co 2 MnIn, shown in Figure 3 , is similar to Co 2 MnSn in the sense that the interplay between the positions of the Co-3 d and Mn-3 d states leads to inconsistent results from different electronic structure methods. With PBE there is a very small gap of 0.103 eV in the minority-spin channel. The lower edge of this gap lies approximately 0.627 eV above the Fermi level. Similar to Co 2 MnSn, there are contributions from both Co and Mn d states near the Fermi level, resulting in a spin polarization of 86.0%. With HSE, the Mn d states are strongly exchange-split and pushed away from the Fermi level. The gap of 1.97 eV in the minority spin channel is much larger than with PBE and found 0.997 eV above the Fermi level. The main peaks of the Co d states are also pushed away from the Fermi level, leaving a relatively small minority spin density tail. The In p states contribute to the majority channel at the Fermi level, resulting in a minority spin polarization of -64.4%. With QPGW the gap in the minority spin channel is 0.196 eV and it is located 1.199 eV above the Fermi level. The Mn d states are not near the Fermi level, although not as far away as with HSE. The DOS at the Fermi level is dominated by a large peak of minority spin Co d states, yielding a high spin polarization of -94.6%. PBE+U(BO) produces a gap of 0.669 eV in the minority spin channel, which lies 0.499 eV above the Fermi level. The relative positions of the Co 3 d and Mn 3 d states is similar to QPGW with the DOS around the Fermi level dominated by a large Co 3 d minority spin peak. Unlike pure PBE, which favors majority-spin polarization, PBE+U(BO) yields the same minority-spin polarization of -94.3% as QPGW. If we consider the results of QPGW to be the most reliable, then Co 2 MnIn is expected to be a near-half-metal with a high spin polarization and a high density of states at the Fermi level, which would be advantageous for spintronic applications. Figure 4: Band structure and element-resolved DOS of Co 2 TiSn calculated using (a) PBE, (b) HSE, (c) QPGW, and (d) PBE+U(BO). In the band-structure panels, the majority and minority spin channels are shown in blue and red, respectively. In the DOS panels, the majority and minority spin components are plotted on the left and right sides, respectively. The rightmost panels present a magnified view of the DOS around the Fermi level. The total magnetic moment (MM, in \mu B ) and spin polarization percentage at the Fermi level (SP%) are listed for each method. As mentioned above, Co 2 TiSn, which has long been regarded as a prototypical example of a half-metallic Heusler compound Miura et al. 2006 ; Lee et al. 2005 ; Hickey et al. 2006 , is the only material unanimously predicted by all methods used here to be a half-metal. As shown in Figure 4 , there are still significant differences between the band structures produced by different methods with respect to the size of the gap in the minority spin channel and its position in energy and in k -space. PBE yields a gap of 0.605 eV, which is significantly smaller than the values of 1.291 eV, 1.563 eV, and 1.016 eV produced by HSE, QPGW, and PBE+U(BO), respectively. In addition, with PBE and QPGW the gap is located at the \Gamma -point, whereas with HSE and PBE+U(BO) it is located at the L-point. These differences in the band structure do not affect the outcome that only the majority spin Co d states contribute to the DOS at the Fermi level, which results in a 100% spin polarization. Experimentally, however, spin polarization values obtained via point-contact Andreev reflection (PCAR) are consistently lower, with reported values of 64 \pm 2%, 57%, and 58.9 \pm 2.2% Bainsla and Suresh 2016 ; Varaprasad et al. 2012 ; Ooka et al. 2017 . A possible explanation is that the spin polarization measured by PCAR is highly sensitive to perturbations such as defects, disorder, surface effects, or deviations from stoichiometry Klaer et al. 2011 , even when the minority-spin gap edge is not located particularly close to the Fermi level. Figure 5: Band structure and element-resolved DOS of Co 2 ZrAl calculated using (a) PBE, (b) HSE, (c) QPGW, and (d) PBE+U(BO). In the band-structure panels, the majority and minority spin channels are shown in blue and red, respectively. In the DOS panels, the majority and minority spin components are plotted on the left and right sides, respectively. The rightmost panels present a magnified view of the DOS around the Fermi level. The total magnetic moment (MM, in \mu B ) and spin polarization percentage at the Fermi level (SP%) are listed for each method. Negative spin polarization indicates that the minority spin channel is dominant around the Fermi level. Co 2 ZrAl is predicted to be a half-metal by PBE, QPGW, and PBE+U(BO). With all three methods, only the majority spin Co d states contribute to the DOS at the Fermi level, leading to a 100% spin polarization, as shown in Figure 5 . However, these three methods differ in the details of the band structure, particularly in the magnitude and location in k -space of the minority-spin gap. With PBE, the minority-spin gap is 0.515 eV, whereas QPGW predicts a substantially larger gap of 1.913 eV. The PBE+U(BO) result of 1.044 eV lies between these two values. In addition, with PBE and PBE+U(BO) the minority-spin gap is indirect between the X and \Gamma points, whereas with QPGW it is direct at the X-point. HSE produces a markedly different picture. The majority spin Co d states are pushed away from the Fermi level. As a result, the DOS at the Fermi level is dominated by minority-spin Co d states, with additional contributions from majority-spin Zr d states and only a marginal contribution of Al p states, leading to an overall minority spin polarization of -32.8%. IV Conclusion In summary, we investigated the electronic structure and magnetic properties of six Heusler compounds in the cubic L2_{1} ( Fm \bar{3} m ) phase with a stable ferromagnetic ground state that are lattice-matched to InAs and GaSb. We compared the results obtained using PBE, HSE, QPGW, and PBE+U(BO) with U values optimized to reproduce as closely as possible the QPGW band structure and magnetic moments. The results reveal a strong dependence of the magnitude and position of the gap in the minority spin channel, as well as the spin polarization at the Fermi level, on the electronic structure method used. In some cases, the predictions of different methods even differ qualitatively with respect to whether a material is expected to be a half-metal and the dominant spin channel (majority or minority) at the Fermi level. Our analysis shows that these differences arise from the treatment of d states with some systematic trends among methods. PBE tends to underestimate, whereas HSE tends to overestimate, the bandwidths and exchange splitting of the d manifolds, particularly in materials with strong d – d hybridization. The results of QPGW tend to be in the range between PBE and HSE. In most cases, PBE+U(BO) reproduces well the salient features of the QPGW band structure. The only exceptions found here are Co 2 MnSn and Ni 2 MnSb, for which the spin polarization at the Fermi level obtained with PBE+U(BO) differs qualitatively from QPGW. This may occur because the spin polarization at the Fermi level is not included in the BO objective function. Of the materials considered here, we expect Co 2 TiSn and Co 2 ZrAl to be half-metals with a high degree of confidence, based on consistent predictions of all or most of the methods used here. In addition, if we consider the results of QPGW to be the most reliable, then Co 2 MnIn is expected to be a near-half-metal with a high minority spin polarization at the Fermi level. It is also interesting to note that Co 2 MnIn is predicted to have a high DOS at the Fermi level, in contrast to Co 2 TiSn and Co 2 ZrAl. A high DOS at the Fermi level could be an advantage for spintronic devices because more charge carriers would be available to conduct current and because the spin polarization would be more robust to small perturbations. Our findings have significant implications for the reliability of high-throughput studies of magnetic Heuslers and related materials based on semi-local DFT. We recommend further screening of candidate materials using more accurate methods prior to pursuing them experimentally. For simulations of interfaces and other large systems involving Heuslers, we suggest PBE+U(BO) as a practical solution for obtaining qualitatively correct results at a reasonable computational cost. However, we recommend verifying the results of PBE+U(BO) on a case by case basis and adapting the objective function if needed. This study also highlights the critical need for probing the band structure of Heuslers and related materials by ARPES, preferably spin resolved, in order to assess the accuracy and reliability of electronic structure methods for this important class of materials. V