Simulating superconductivity in mixed-dimensional $t_\parallel$-${J}_\parallel$-${J}_\perp$ bilayers with neural quantum states

Using neural quantum states to simulate a mixed-dimensional $t_\parallel$-$J_\parallel$-$J_\perp$ model, this study provides the first high-precision numerical evidence for superconductivity in two-dimensional bilayers, identifying a crossover between BEC and BCS pairing regimes and a transition between interlayer $s$-wave and intralayer $d$-wave symmetries.

The Gist

The Tale of the Two-Story Dance Floor: Understanding Superconductivity

Imagine you are at a massive, two-story music festival. On each floor, there are thousands of dancers (these are the electrons). Usually, these dancers are a bit chaotic—they bump into each other, move randomly, and create a lot of "friction" (which scientists call electrical resistance).

But under very specific conditions, something magical happens: the dancers stop bumping into each other and start moving in perfect, synchronized pairs. When these pairs move together, they glide through the crowd without hitting anyone. This is superconductivity—the ability for electricity to flow with zero wasted energy.

This paper is about a specific, high-tech "dance floor" called a bilayer nickelate, and the researchers used a "super-computer brain" to figure out exactly how the dancers pair up.

---

1. The Setting: The Two-Story Dance Floor (The Bilayer)

The researchers aren't looking at a single flat floor; they are looking at a bilayer. Imagine two parallel floors (Layer 0 and Layer 1) with a small gap between them.

In this specific model (the mixD model), the dancers have a unique rule: they can dance side-to-side on their own floor very easily, but they can only move between the floors by "jumping" or interacting through a special magnetic force. This "mixed" way of moving is what makes the physics so strange and interesting.

2. The Mystery: How do they pair up?

The big question is: How do these dancers form pairs? The researchers found two main ways this happens, depending on how "strong" the connection is between the two floors.

The "Bose-Einstein" Style (The Tight Huggers)

Imagine the connection between the floors is incredibly strong. The dancers on the top floor and the bottom floor are so attracted to each other that they immediately grab a partner from the opposite floor and form a tight, inseparable duo.

• The Analogy: It’s like a ballroom dance where every couple is locked in a permanent, tight hug. They move as single, solid units (called BEC pairs). They don't care much about the floor; they only care about their partner.

The "BCS" Style (The Social Butterflies)

Now, imagine the connection between the floors is much weaker. The dancers aren't glued to a specific partner on the other floor. Instead, they feel a general "vibe" or attraction to many people at once.

• The Analogy: This is more like a massive, flowing line dance. The dancers are still paired up, but the pairs are "loose" and spread out across the room. They are more aware of the whole dance floor (called BCS pairs).

The Discovery: The researchers proved that by simply changing the "strength" of the connection between the floors, you can watch the dancers transition from "Tight Huggers" to "Social Butterflies."

3. The Plot Twist: Changing the Dance Style (Symmetry)

The researchers also found that if you change the "music" (the internal magnetic forces), the dancers change their entire style:

• Interlayer s-wave: The dancers pair up vertically (one on top, one on bottom).
• Intralayer d-wave: The dancers decide to stay on the same floor but dance in a complex, star-shaped pattern with their neighbors.

The paper shows that there is a "sharp line" where the dancers suddenly switch from one style to the other.

4. The Tool: The "Neural Quantum State" (The AI Choreographer)

How did they simulate this? This is too complex for a normal calculator. They used Neural Quantum States (NQS).

Think of NQS as an AI Choreographer. Instead of trying to track every single dancer's movement manually (which would take a billion years), they trained an Artificial Intelligence to "learn" the rhythm of the dance. The AI looks at the crowd and says, "Based on the music and the floor, this is the most likely way the dancers are moving."

Because this AI is so smart, it can simulate much larger "dance floors" (up to 8x8x2 sites) than previous methods, giving us a much clearer picture of the real world.

Why does this matter?

By understanding these "dance moves," scientists are getting closer to designing new materials that can conduct electricity perfectly at higher temperatures. This could lead to super-fast computers, incredibly efficient power grids, and even hovering Maglev trains that use almost no energy.

Technical Summary

Technical Summary: Simulating Superconductivity in Mixed-Dimensional $t_{\parallel}$-$J_{\parallel}$-$J_{\perp}$ Bilayers with Neural Quantum States

1. Problem Statement

The study is motivated by the 2023 discovery of high-temperature superconductivity ($T_c = 80$ K) in the bilayer nickelate $\text{La}_3\text{Ni}_2\text{O}_7$ (LNO) under pressure. The physics of LNO is believed to be governed by strong electronic correlations involving $d_{z^2}$ and $d_{x^2-y^2}$ orbitals. A proposed effective low-energy description is the mixed-dimensional (mixD) $t_{\parallel}$-$J_{\parallel}$-$J_{\perp}$ bilayer model.

In this model, hopping is restricted to the layers ($t_{\parallel}$), while magnetic exchange occurs both within layers ($J_{\parallel}$) and between layers ($J_{\perp}$). While previous studies used Matrix Product States (MPS), they were largely limited to quasi-1D geometries (ladders or coupled ladders). A fully two-dimensional (2D) numerical treatment that captures strong correlation effects beyond mean-field theory has been a significant challenge in the field.

2. Methodology

The authors employ Neural Quantum States (NQS), a variational approach using deep learning to represent many-body wave functions. Specifically, they introduce and adapt the Gutzwiller-projected Hidden Fermion Pfaffian State (G-HFPS).

• Wave Function Ansatz: The G-HFPS combines the Hidden Fermion Pfaffian State (HFPS)—which naturally encodes charge pairing through an expanded Hilbert space of "hidden" fermions—with Gutzwiller projection to explicitly enforce the strong local correlation (no double occupancy) inherent in $t$-$J$ models.
• Neural Network Architecture: They utilize a Group Convolutional Neural Network (GCNN). This allows the wave function to inherently respect the lattice symmetries (translation, rotation, reflection, and spin-parity), significantly improving computational efficiency and accuracy.
• Optimization: The parameters are optimized using the stochastic reconfiguration algorithm. To explore the phase diagram, they use adiabatic transport, a protocol that evolves the wave function continuously as Hamiltonian parameters are tuned, ensuring the simulation stays in the ground state.
• Observables: To detect superconductivity in finite-size systems with conserved particle numbers, they use a mixed momentum-real space estimator $\langle \hat{P}_{\Gamma}(k) \rangle$. This allows them to probe the pairing order parameter and the spatial extent of Cooper pairs.

3. Key Contributions

• First 2D NQS Simulation of a Fermionic Bilayer: This represents the first time NQS has been applied to a multi-orbital fermionic bilayer system to study 2D physics.
• High-Precision Numerics: The study provides the first high-precision numerical evidence for superconductivity in these 2D bilayers, validated by comparing NQS results with MPS results on coupled ladders.
• Comprehensive Phase Diagram Mapping: The work maps out the interplay between interlayer and intralayer exchange and doping, identifying critical crossovers and transitions.

4. Results

The simulations reveal a rich phase diagram characterized by two major phenomena:

• BEC-to-BCS Crossover: By tuning the ratio of intralayer hopping to interlayer exchange ($t_{\parallel}/J_{\perp}$), the authors observe a crossover in pairing behavior.
  • At strong $J_{\perp}$, the system exhibits Bose-Einstein Condensation (BEC) of tightly bound, local interlayer pairs.
  • As $t_{\parallel}$ increases, the pairs become more spatially extended, transitioning into a Bardeen-Cooper-Schrieffer (BCS) regime of long-range Cooper pairs.
• Pairing Symmetry Transition: By tuning the intralayer exchange $J_{\parallel}$, the authors identify a sharp, first-order transition in the pairing symmetry:
  • Interlayer $s$-wave ($s_{\perp}$) pairing dominates when $J_{\parallel}$ is small.
  • Intralayer $d$-wave ($d_{\parallel}$) pairing emerges when $J_{\parallel}$ exceeds a critical value ($J_{\parallel} \approx t_{\parallel}$).

5. Significance

This research provides crucial theoretical insights into the microscopic mechanism of superconductivity in bilayer nickelates like $\text{La}_3\text{Ni}_2\text{O}_7$. It supports the hypothesis that the pairing mechanism is tied to the competition between different exchange scales and potentially involves a Feshbach-like resonance between BEC and BCS channels. Furthermore, the success of the G-HFPS method demonstrates that NQS is a powerful, scalable framework for simulating complex, strongly correlated quantum materials that are currently inaccessible to traditional numerical methods. This has direct implications for both condensed matter physics and the development of cold atom quantum simulation platforms.