Add refined abstracts to publication pages

Author: Karasawa-Shin

_publications/2023-04-01-deep-learning-magnetic-superstructures.md

@@ -9,3 +9,5 @@ venue: "Nat. Comput. Sci. <b>3</b>, 321-327 (2023) [Cover story]"
 authors: "He Li*, <b>Zechen Tang</b>*, ... , Wenhui Duan, and Yong Xu"
 paperurl: 'https://www.nature.com/articles/s43588-023-00424-3'
 ---
+
+Magnetic superstructures — from spin spirals to skyrmions — host rich quantum phenomena but are notoriously hard to simulate with first-principles methods. We extended the DeepH framework to **magnetic materials** by designing neural networks that rigorously respect Euclidean and time-reversal symmetries, and by learning from diverse magnetic configurations generated via constrained DFT. The method (**xDeepH**) successfully predicts electronic structures of magnetic skyrmions in moiré-twisted CrI₃, revealing how skyrmions can suppress flat bands — a coupling previously inaccessible to ab initio studies.

_publications/2024-03-01-deep-learning-dfpt.md

@@ -9,3 +9,5 @@ venue: "Phys. Rev. Lett. <b>132</b>, 096401 (2024) [Editors&apos; suggestion]"
 authors: "He Li*, <b>Zechen Tang</b>*, ... , Wenhui Duan, and Yong Xu"
 paperurl: 'https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.132.096401'
 ---
+
+Calculating materials' response to external perturbations — such as electron-phonon coupling — is essential for understanding superconductivity and transport, yet traditional density functional perturbation theory (DFPT) is computationally prohibitive for large systems. We showed that **neural networks can emulate DFPT** by combining deep-learning DFT Hamiltonians with automatic differentiation, achieving up to **three orders of magnitude speedup** while preserving ab initio accuracy. This work unifies ground-state DFT and DFPT under a single deep-learning framework.

_publications/2024-07-01-universal-materials-model.md

@@ -9,3 +9,5 @@ venue: "Sci. Bull. <b>69</b>, 2514 (2024) [Cover story]"
 authors: "Yuxiang Wang*, Yang Li*, <b>Zechen Tang</b>*, ... , Wenhui Duan, and Yong Xu"
 paperurl: 'https://www.sciencedirect.com/science/article/pii/S2095927324004079'
 ---
+
+Large materials models — analogous to large language models but for materials science — could revolutionize materials discovery, but building one that handles arbitrary compositions and structures is an open challenge. We demonstrated a concrete path forward: by constructing a diverse materials database and substantially improving the DeepH method, we trained a **universal DeepH model** that accurately predicts DFT Hamiltonians across diverse elemental compositions and crystal structures. We further showed how this universal model can be **fine-tuned** for specialized applications, laying the groundwork toward large materials models.

_publications/2024-08-01-neural-nn-dft-variational.md

@@ -9,3 +9,5 @@ venue: "Phys. Rev. Lett. <b>133</b>, 076401 (2024) [Editors&apos; suggestion]"
 authors: "Yang Li*, <b>Zechen Tang</b>*, Zezhou Chen*, ... , Wenhui Duan, and Yong Xu"
 paperurl: 'https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.133.076401'
 ---
+
+Conventional deep-learning DFT relies on supervised learning from labeled data, which separates neural network training from the physics of DFT. We flipped this paradigm: **physical principles become the loss function**. By backpropagating through a differentiable DFT code, the neural network learns to represent Hamiltonians without any training labels — achieving higher accuracy than supervised approaches. This physics-informed, unsupervised framework opens a new direction for developing deep-learning electronic structure methods.

_publications/2024-10-01-deep-hybrid-dft.md

@@ -9,3 +9,5 @@ venue: "Nat. Commun. <b>15</b>, 8815 (2024)"
 authors: "<b>Zechen Tang</b>*, He Li*, Peize Lin*, ... , Xinguo Ren, Wenhui Duan, and Yong Xu"
 paperurl: 'https://www.nature.com/articles/s41467-024-53028-4'
 ---
+
+Hybrid density functional theory (DFT) fixes the notorious "band gap problem" of standard DFT by incorporating exact exchange, but at a daunting computational cost. We developed **DeepH-hybrid**, an equivariant neural network that learns the hybrid-functional Hamiltonian directly from material structure, bypassing costly self-consistent iterations. Trained on small structures, the model generalizes to large supercells — including **magic-angle twisted bilayer graphene** with over 11,000 atoms — making hybrid-DFT accuracy affordable for large-scale materials simulations for the first time.

_publications/2025-01-01-deep-learning-electronic-structure-review.md

@@ -9,3 +9,5 @@ venue: "Nat. Comput. Sci. <b>5</b>, 1133 (2025) [Review]"
 authors: "<b>Zechen Tang</b>*, Haoxiang Chen*, ..., Wenhui Duan, Ji Chen, and Yong Xu"
 paperurl: 'https://www.nature.com/articles/s43588-025-00932-4'
 ---
+
+First-principles electronic structure calculations face a fundamental **accuracy–efficiency dilemma**: highly accurate quantum Monte Carlo methods are too expensive for large systems, while efficient DFT sacrifices precision. This review surveys two emerging deep-learning paradigms that are breaking through this bottleneck: **deep-learning quantum Monte Carlo** (DL-QMC) achieves unprecedented accuracy by using neural networks as many-body wavefunction ansatzes, while **deep-learning DFT** (DL-DFT) enables million-atom simulations with ab initio quality. We discuss the key methodological advances and envision how the synergy of these approaches may reshape the future of computational materials science.

_publications/2025-02-01-nobel-prize-physics-2024.md

@@ -9,3 +9,5 @@ venue: "PHYSICS (Wuli) <b>54</b>, 1 (2025)"
 authors: "<b>Zechen Tang</b>, Wenhui Duan, and Yong Xu"
 paperurl: 'https://wuli.iphy.ac.cn/en/article/cstr/32040.14.wl20250101'
 ---
+
+The 2024 Nobel Prize in Physics honored John Hopfield and Geoffrey Hinton for foundational discoveries that enabled today's deep learning revolution. Written for a broad physics audience, this popular science article traces the deep intellectual connections between physics and neural networks — from statistical mechanics inspiration to modern applications in first-principles materials simulations — and offers a glimpse into how deep learning is reshaping the way physicists understand and predict the quantum world.