---
title: MACE-H Graph Neural Network for Hamiltonian Prediction
url: https://www.emergentmind.com/topics/mace-h-graph-neural-network
type: topic
---

# MACE-H Graph Neural Network for Hamiltonian Prediction

to արտաքին? No, final only.
Searching arXiv for the cited papers to ground the article in the current literature.
MACE-H is an $O(3)$-equivariant many-body graph neural network for predicting Kohn–Sham density functional theory Hamiltonians from atomistic structure. In the formulation introduced in “Equivariant Electronic Hamiltonian Prediction with Many-Body Message Passing,” MACE-H combines high body-order message passing with a node-order expansion so that the model can generate all relevant $O(3)$ irreducible representations needed for block-structured Hamiltonians up to $f$ orbitals, including on-site and inter-site interactions and, where relevant, spin-resolved channels under spin–orbit coupling (SOC) [2508.15108]. The name also sits within a broader lineage: MACE itself originates from higher-order equivariant message passing in interatomic force-field learning [2206.07697], while a later review of GAP, ACE, and MACE noted that “MACE-H” was not yet a separately defined architecture in that context and instead connected the label to hierarchical or multihead stabilizing modifications used during fine-tuning [2410.06354]. As a result, the term has both a precise meaning in Hamiltonian learning and an earlier, looser usage in the wider MACE literature.

## 1. Nomenclature and historical placement

The immediate antecedent of MACE-H is MACE, introduced as a “Higher Order Equivariant Message Passing Neural Network” for fast and accurate force fields [2206.07697]. That work argued that most equivariant message passing neural networks rely on two-body message passing, so higher-body correlations must be built indirectly through depth and nonlinearities. MACE addressed this by constructing higher body-order messages, specifically showing that “using four-body messages reduces the required number of message passing iterations to just two” [2206.07697]. In that original force-field setting, however, the term “MACE-H” was not defined as a separate model.

A later review, “From GAP to ACE to MACE,” placed MACE within the evolution from SOAP–GAP to ACE and then to equivariant message-passing models, but it explicitly stated that the review “does not introduce or define ‘MACE-H’ as a separate architecture” [2410.06354]. In that review’s usage, the closest related idea was “fine-tuning (with multihead stabilization),” together with broader hierarchical or hybrid design choices.

The specific architecture now called MACE-H appears in the Hamiltonian-learning literature. In “Equivariant Electronic Hamiltonian Prediction with Many-Body Message Passing,” MACE-H is defined as a graph neural network that extends the Many-body Atomic Cluster Expansion paradigm from energies and forces to electronic Hamiltonians [2508.15108]. This usage is distinct from ACE-HGNN, the “Adaptive Curvature Exploration Hyperbolic Graph Neural Network,” which is a multi-agent reinforcement-learning framework for hyperbolic representation learning and does not use the string “MACE-H” in the paper itself [2110.07888].

This naming history matters because it prevents a common misconception: MACE-H is not a single universally fixed acronym across all graph-learning subfields. In electronic-structure prediction, it denotes a concrete Hamiltonian model; in earlier MACE discussions, it could only be inferred as shorthand for higher-order, hierarchical, or multihead variants.

## 2. Architectural definition in Hamiltonian learning

In its explicit 2025 form, MACE-H is designed to predict Hamiltonian matrix elements $H_{ij}$ between localized orbitals, including on-site blocks $(i=j)$ and inter-site blocks $(i\neq j)$, with orbital shells spanning $s$ $(\ell=0)$, $p$ $(\ell=1)$, $d$ $(\ell=2)$, and $f$ $(\ell=3)$ channels [2508.15108]. The core architectural statement is that it “combines high body-order message passing with a node-order expansion to efficiently obtain all relevant $O(3)$ irreducible representations” [2508.15108].

The graph construction is standard for atomistic GNNs. Nodes are atoms; edges connect neighbors within a radial cutoff; inputs include atomic species $Z$, positions $\{r_i\}$, lattice vectors, and periodic boundary conditions. Node features carry species embeddings and equivariant hidden states arranged by irreducible representations up to a chosen $\ell_{\max}$, while edge features decompose relative geometry into radial basis functions and spherical harmonics $Y_{\ell m}(\hat r_{ij})$ [2508.15108].

For each node $i$ and angular momentum $\ell$, equivariant features are tensors $F_i^{(\ell)}$ with components $F_{i,c,m}^{(\ell)}$. On an edge $(i,j)$, the model uses a radial–angular decomposition
$$
E^{(\ell)}_{ij,c,m} = \sum_n \alpha^{(\ell)}_{c,n}(Z_i,Z_j)\,\phi_n(r_{ij})\,Y_{\ell m}(\hat r_{ij}),
$$
with learnable coefficients $\alpha^{(\ell)}_{c,n}$ that may be species-dependent [2508.15108]. Equivariant couplings are implemented through Clebsch–Gordan decomposition, so tensor products of irreps are projected into target irreps with the correct $O(3)$ transformation law.

A prototypical update is written as
$$
F^{(\ell,t+1)}_{i,m}
=
\sum_{j\in\mathcal N(i)}
\sum_{\ell_1,\ell_2}
\sum_{m_1,m_2}
W^{(t)}_{\ell_1,\ell_2\to\ell}
C^{\ell m}_{\ell_1 m_1,\ell_2 m_2}
F^{(\ell_1,t)}_{i,m_1}
E^{(\ell_2)}_{ij,m_2},
$$
followed by an equivariant nonlinearity [2508.15108]. Because the outputs are Hamiltonian blocks rather than scalar energies, equivariance is not a decorative property but a structural requirement: $p$, $d$, and $f$ orbital blocks are intrinsically direction-sensitive.

The output stage maps equivariant node and edge features to shell-resolved blocks $H_{ij}[L_i,L_j]$. Hermiticity is imposed at assembly time through $H_{ji}=H_{ij}^\dagger$, and on-site blocks are symmetrized [2508.15108]. For periodic systems, real-space blocks $H(R)$ are Fourier-summed into
$$
H(k)=\sum_R H(R)e^{ik\cdot R},
$$
which enables band-structure and density-of-states calculations from predicted matrix elements [2508.15108].

## 3. High body-order message passing and node-order expansion

The many-body component of MACE-H is inherited conceptually from MACE. The original MACE paper framed its central advance as the use of higher body-order messages so that body order is no longer tied to network depth; by constructing four-body messages, MACE reaches high expressivity with only two message-passing iterations [2206.07697]. The later GAP–ACE–MACE review likewise described MACE as feeding ACE’s equivariant and invariant tensor bases into a message-passing graph neural network that is explicitly $SO(3)$-equivariant [2410.06354].

MACE-H adapts this logic to Hamiltonian prediction. High body-order message passing is defined through symmetric polynomials over neighbor contributions, projected to target angular momenta by Clebsch–Gordan decomposition:
$$
P_i^{(\ell,k)}
=
\mathcal D_\ell
\big[
\rho_i^{(\ell_1)}(j_1)\otimes
\rho_i^{(\ell_2)}(j_2)\otimes
\cdots \otimes
\rho_i^{(\ell_k)}(j_k)
\big],
$$
with $1\le k\le \nu$ [2508.15108]. These terms create effective $k$-body features without naïve explicit enumeration of all triplets or quadruplets.

The node-order expansion is the second defining ingredient. To ensure that hidden features can parameterize Hamiltonian blocks up to $f$ orbitals, MACE-H forms higher-order irreps by tensor products of node features followed by Clebsch–Gordan projection:
$$
F_i^{(\ell)}
\leftarrow
\bigoplus_k
\mathcal D_\ell
\big[
F_i^{(\ell_1)}\otimes
F_i^{(\ell_2)}\otimes
\cdots \otimes
F_i^{(\ell_k)}
\big].
$$
This “bootstraps richer angular content from learned lower-$\ell$ channels” and is the mechanism by which the model can efficiently obtain all relevant irreps for $s/p/d/f$ Hamiltonian blocks [2508.15108].

A plausible implication is that the phrase “MACE-H” in the Hamiltonian context can be read as an extension of the original “higher order” idea from force-field MACE into a representation space tailored to orbital-block prediction. That interpretation is consistent with the historical MACE emphasis on body order [2206.07697], but the Hamiltonian paper’s distinctive addition is the node-order expansion [2508.15108].

## 4. Physical constraints, training objective, and stabilization

MACE-H is trained primarily on matrix-element prediction. A canonical multi-task objective given in the Hamiltonian paper is
$$
\mathcal L
=
\lambda_H \|H^{\mathrm{pred}}-H^{\mathrm{DFT}}\|_2^2
+
\lambda_\varepsilon \|\varepsilon^{\mathrm{pred}}-\varepsilon^{\mathrm{DFT}}\|_2^2
+
\Omega,
$$
where $\varepsilon$ denotes eigenvalues and $\Omega$ collects regularization and symmetry penalties [2508.15108]. In the reported experiments, the primary metrics are mean absolute errors of matrix elements, while eigenvalues are validated post hoc for bulk Au [2508.15108].

The paper emphasizes a practical difficulty specific to Hamiltonian learning under SOC: different element–orbital pair sub-blocks can vary in magnitude by orders, which causes gradient instability under naïve scaling. To address this, MACE-H introduces a “shift-and-scale” operation for the last-layer irreps,
$$
o_{ij,klm}
=
o^{(t)}_{ij,klm}
+
\big(o^{(t)}_{ij,klm}(\sigma_{Z_{i,j},k}-1)\big).\mathrm{detach}()
+
\mu_{Z_{i,j},k},
$$
where $\mu$ and $\sigma$ are precomputed means and standard deviations for each sub-block type, and the detach operation prevents the scaling factors from amplifying backpropagated gradients [2508.15108]. The reported effect is a reduction in mean absolute error and improved convergence, especially for small-magnitude cross-spin blocks.

Several physical constraints are built into the workflow rather than left entirely to optimization. Hermiticity is enforced by symmetrization; time-reversal symmetry and permutation invariance are respected by construction of features and aggregation; periodic assembly is handled through real-space-to-$k$-space Fourier summation [2508.15108]. For bulk Au, the paper further reports that core-state projection and sparsification beyond $10$ Å produce negligible effects on eigenvalues, with eigenvalue error $1.41\times10^{-4}$ eV and entropy error $6.96\times10^{-12}$ eV K$^{-1}$ Å$^{-3}$ relative to model error [2508.15108].

The broader MACE review provides context for why these stabilization devices are noteworthy. In that review, multihead stabilization was highlighted as a practically important modification during fine-tuning of foundation models, intended to improve optimization robustness [2410.06354]. Although that is not the same mechanism as the Hamiltonian paper’s shift-and-scale operation, both address the same general issue: training equivariant many-body models can be limited by optimization stability rather than representational capacity alone.

## 5. Datasets, reported performance, and empirical behavior

The Hamiltonian paper evaluates MACE-H on several open materials benchmark datasets and a new dataset for bulk gold [2508.15108]. The 2D datasets include monolayer graphene, monolayer MoS$_2$, bilayer graphene, bilayer Bismuthene, bilayer Bi$_2$Se$_3$, and bilayer Bi$_2$Te$_3$, with SOC included for Bismuthene, Bi$_2$Se$_3$, and one Bi$_2$Te$_3$ setting [2508.15108]. Bulk Au is computed with FHI-aims and treated without SOC [2508.15108].

The paper reports the following representative matrix-element MAEs in meV:

| System | Baseline(s) | MACE-H |
|---|---|---|
| Graphene (noSOC) monolayer | DeepH 2.1; DeepH-E3 0.27 | 0.21 |
| MoS$_2$ (noSOC) monolayer | DeepH 0.95; DeepH-E3 0.45 | 0.39 |
| Bilayer graphene (noSOC), shifted/twisted | DeepH-E3 0.40/0.28 | 0.31/0.24 |
| Bilayer Bi$_2$Se$_3$ (SOC), shifted/twisted | DeepH-E3 0.40/0.35 | 0.34/0.35 |
| Bulk Au (noSOC) | DeepH-E3 0.41 (max 0.80) | 0.32 (max 0.65) |
| Bulk Au + shift-and-scale | — | 0.27 (max 0.62) |

Across monolayers and shifted bilayers, MACE-H is reported to outperform DeepH-E3 and non-equivariant DeepH, with overall MAEs “down to $0.21$–$0.40$ meV” [2508.15108]. The same paper also reports that MACE-H “consistently improves in-distribution shifted bilayers,” whereas “out-of-distribution twisted bilayers are more challenging for high $\nu$ and high $L_{\max}$” [2508.15108]. This is not a contradiction; it reflects a stated locality preference.

For bulk Au, the reference model uses $T=2$ layers, $\nu=2$, $L_{\mathrm{hidden}}=2$, $L_{\mathrm{edge}}=4$, $L_{\mathrm{MACE}}=5$, learning rate $0.008$, and batch size $2$ [2508.15108]. For monolayer graphene, an example configuration is learning rate $0.006$, batch size $1$, $3$ layers, $\nu=3$, and $L_{\mathrm{MACE}}=L_{\mathrm{hidden}}=L_{\mathrm{edge}}=5$ [2508.15108]. These concrete settings indicate that the model is not tied to a single hyperparameter regime.

The original MACE paper provides a complementary empirical baseline for the higher-order design principle. On rMD17, 3BPA, and AcAc, MACE reached or exceeded state-of-the-art accuracy while remaining substantially faster than NequIP or BOTNet at comparable accuracy in several settings [2206.07697]. This does not establish Hamiltonian performance directly, but it clarifies why higher body-order message passing was a natural foundation for extending MACE toward more demanding equivariant targets.

## 6. Computational properties, limitations, and relation to adjacent models

The Hamiltonian paper states that message passing per layer scales roughly as
$$
O\!\left(N n_{\mathcal N}\cdot \sum_\ell C_\ell^2\right),
$$
where $N$ is the number of atoms, $n_{\mathcal N}$ the number of neighbors per atom, and $C_\ell$ the number of channels for irrep $\ell$ [2508.15108]. Many-body terms add polynomial factors up to order $\nu$, but these are implemented via tensor factorizations and Clebsch–Gordan projections so as to avoid the combinatorial blow-up of naïve $n$-body enumeration [2508.15108]. In the reported settings, inference time on a single A100 GPU is comparable to DeepH-E3 because the dominant cost lies in the shared edge-update block, while the many-body expansion and node-order expansion add negligible overhead [2508.15108].

The same work identifies a central trade-off between locality and transferability. Higher $\nu$ and higher $L_{\max}$ improve local shifted-bilayer predictions but can worsen twisted-bilayer out-of-distribution performance; for twisted systems, the recommendation is to reduce $\nu$ and $L_{\max}$ and enlarge the training set [2508.15108]. Perturbation-response analysis further shows that MACE-H exhibits “faster attenuation with distance,” which the authors interpret as consistent with its locality preference [2508.15108].

This locality issue aligns with a broader limitation already recognized in MACE and ACE more generally. The 2024 review noted that short-range ACE/MACE frameworks do not by themselves solve long-range electrostatics or charge transfer, and that improved scaling and parallelism remain open engineering directions [2410.06354]. The Hamiltonian paper formulates the corresponding limitation in electronic-structure language: long-range interactions beyond the cutoff may be underrepresented, and nonlocal terms may be needed for moiré or twist systems [2508.15108].

Comparison with neighboring architectures is therefore best made along two axes. Relative to DeepH and DeepH-E3, MACE-H adds explicit many-body and node-order machinery while preserving $O(3)$ equivariance [2508.15108]. Relative to the original force-field MACE, it changes the target from scalar site energies and derived forces to block-structured Hamiltonians with shell-resolved angular content [2206.07697]. A common misconception is to treat MACE-H as merely “MACE plus a new output head.” The Hamiltonian paper suggests a stronger claim: the node-order expansion is specifically introduced to generate all irreps needed for $f$-orbital matrix interaction blocks, which is a target-space requirement absent from standard interatomic-potential formulations [2508.15108].

A final source of confusion comes from unrelated terminology in graph learning. ACE-HGNN is a hyperbolic graph neural network with adaptive curvature exploration via multi-agent reinforcement learning [2110.07888]. Despite superficial overlap in initials, it belongs to a different methodological lineage and addresses a different problem.

## 7. Significance and scope of current usage

Within electronic-structure machine learning, MACE-H occupies the intersection of equivariant GNNs, ACE-style many-body expansions, and Hamiltonian surrogate modeling. Its stated objective is to enable rapid prediction of band structures and densities of states without repeated self-consistent DFT, by learning localized Hamiltonian blocks and then assembling periodic Hamiltonians in $k$-space [2508.15108]. The reported outcome is “sub-meV prediction errors on matrix elements and eigenvalues across all systems,” together with computational efficiency suitable for high-throughput screening [2508.15108].

In the longer genealogy of atomistic machine learning, MACE-H can be viewed as a specialization of the MACE design philosophy rather than a replacement of it. GAP supplied symmetry-correct descriptors and probabilistic regression; ACE supplied a systematic symmetry-adapted basis with controllable body order; MACE fused ACE-like equivariant structure with message passing; and MACE-H extends that route to orbital-resolved Hamiltonian prediction [2410.06354]. This suggests a broader pattern: once equivariant many-body features are available, the main remaining question is how to tailor the representation to the physical tensorial object being predicted.

The term “MACE-H Graph Neural Network” therefore has a precise modern referent: an $O(3)$-equivariant many-body GNN for Kohn–Sham Hamiltonians, built from high body-order message passing and node-order expansion [2508.15108]. At the same time, the historical record shows that earlier MACE literature used adjacent ideas—higher-order message passing, hierarchical aggregation, and multihead stabilization—without defining a separate MACE-H architecture [2206.07697; 2410.06354]. Recognizing that distinction is essential for accurate reading of the recent literature.

Source: https://www.emergentmind.com/topics/mace-h-graph-neural-network