Uncovering Exotic Paired States in the 2D Spin-Imbalanced Fermi Gas with Neural Wave Functions

Using neural network variational Monte Carlo methods, this study maps the zero-temperature phase diagram of a 2D spin-imbalanced Fermi gas, revealing a rich landscape of exotic states including the Fulde-Ferrell-Larkin-Ovchinnikov phase, polarized superfluids, phase separation, and a unique crystalline phase of Cooper pairs embedded in a Fermi fluid.

The Gist

Imagine a crowded dance floor where two types of dancers are present: a large group of "Spin-Up" dancers and a smaller group of "Spin-Down" dancers. The rules of this dance floor are unique: the Spin-Down dancers are attracted to the Spin-Up dancers and want to hold hands, but they don't want to hold hands with their own kind.

This paper is a high-tech simulation of what happens on this dance floor when the music (the interaction strength) changes from a slow, gentle waltz to a frantic, intense rave. The researchers used a super-smart computer brain (a neural network) to figure out exactly how these dancers arrange themselves when the temperature is absolute zero (the coldest possible state).

Here is what they discovered, broken down into three distinct "moods" of the dance floor:

1. The Gentle Waltz (Weak Interactions)

When the attraction between the dancers is weak, the Spin-Down dancers pair up with Spin-Up dancers, but they do it in a very specific, wavy pattern.

• The Discovery: Instead of pairing up right next to each other, the pairs form a pattern where they carry a little bit of "momentum" or motion together.
• The Metaphor: Imagine couples dancing in a circle, but the whole circle is slowly rotating around the room. This is called the FFLO phase. It's like a synchronized dance where the partners are slightly offset, creating a wave-like motion across the floor.

2. The Intense Rave (Strong Interactions)

When the attraction becomes very strong, the Spin-Down dancers cling tightly to the Spin-Up dancers, forming tight little duos (like bosons).

• The Discovery: The dance floor splits into two distinct zones. In one zone, you have the tight Spin-Up/Spin-Down pairs huddled together. In the other zone, the extra Spin-Up dancers (who couldn't find a partner) are left alone, forming a "sea" of unpaired dancers.
• The Metaphor: It's like a party where the popular couples have formed a tight-knit circle in the center, while the single guys are pushed to the edges, forming a ring around them. The single Spin-Up dancers move freely like a fluid, while the pairs are stuck together.
• The "Hole": The researchers noticed something strange in the "momentum" of the Spin-Down dancers. Because the Spin-Up dancers are already occupying the best spots (the center of the dance floor), the Spin-Down dancers are blocked from those spots. It's as if the Spin-Down dancers have a "hole" in their dance map where they simply cannot go because the Spin-Up dancers are already there.

3. The Crystal Formation (The Surprise)

The most surprising discovery happened in the middle ground, when the attraction is just right—not too weak, not too strong.

• The Discovery: The tightly bound pairs stopped moving around randomly and decided to stand still in a perfect, repeating grid pattern. They formed a crystal.
• The Metaphor: Usually, crystals (like ice or salt) form because particles push each other away (repulsion). But here, the particles are attracted to each other! It's as if the couples, by holding hands so tightly, accidentally created an invisible force field that pushes other couples away, forcing them to stand in a perfect, rigid lattice.
• The Scene: Imagine the dance floor where the couples have frozen into a perfect checkerboard pattern, while the unpaired Spin-Up dancers flow around them like water flowing around stones in a river.

How They Did It

The researchers didn't just guess; they used a "Neural Network Variational Monte Carlo" method. Think of this as a super-advanced AI that acts like a million different dancers trying out different arrangements simultaneously. The AI learns which arrangement uses the least amount of energy, effectively finding the most stable "dance formation" for the system.

The Bottom Line

This study reveals that even in a system where everything is attracted to everything else, nature can create complex structures like crystals and phase-separated islands. They found a new, exotic state of matter where fermions (the dancers) spontaneously organize into a crystal lattice, a phenomenon that hadn't been seen before in this specific type of 2D gas. It shows that when you mix different amounts of "spin" and tune the attraction just right, the universe can get very creative with how it arranges its particles.

Technical Summary

1. Problem Statement

The paper addresses the challenge of characterizing the zero-temperature ($T=0$) phase diagram of the two-dimensional spin-imbalanced Fermi gas (2D SIFG) with short-range attractive interactions. This system is a canonical model for studying spin-imbalanced superfluidity, relevant to ultracold atomic gases and high-$T_c$ superconductors.

• The Challenge: The 2D SIFG exhibits complex many-body correlations, particularly in the crossover regime between the Bardeen-Cooper-Schrieffer (BCS) superfluid and the Bose-Einstein condensate (BEC) limits.
• Limitations of Existing Methods:
  • Mean-field theories (e.g., standard BCS) fail to capture strong correlations in 2D.
  • Density Matrix Renormalization Group (DMRG) is restricted to quasi-1D geometries.
  • Conventional Quantum Monte Carlo (QMC) methods suffer from the fermionic sign problem and limited expressivity of trial wave functions (Ansätze).
• Goal: To quantitatively map the ground-state phases of the 2D SIFG across the entire BCS-BEC crossover, specifically investigating the existence of exotic phases like the Fulde-Ferrell-Larkin-Ovchinnikov (FFLO) state, the Sarma phase, and potential crystalline structures.

2. Methodology

The authors employ the Neural Network Variational Monte Carlo (NNVMC) method, a state-of-the-art technique that utilizes deep learning to construct highly expressive variational wave functions.

• Wave Function Ansatz: They utilize the AGPs (Antisymmetrized Geminal Power) FermiNet architecture.
  • This extends the original FermiNet (designed for normal systems) to handle superfluids and spin-imbalanced systems.
  • The wave function $\Psi$ is constructed as a sum of determinants ($D$) of "geminals" (pairing functions) and "orbitals" (for unpaired particles).
  • For a system with $N_\uparrow$ spin-up and $N_\downarrow$ spin-down particles (where $N_\uparrow > N_\downarrow$), the ansatz is:
    $$\Psi = \sum_{k}^{D} \det \begin{pmatrix} \phi_k(r^\uparrow_i, r^\downarrow_j) & \dots & \phi_k(r^\uparrow_i, r^\downarrow_p) & \varphi_{k\uparrow}^1(r^\uparrow_i) & \dots \\ \vdots & \ddots & \vdots & \vdots & \ddots \end{pmatrix}$$
    where $p = N_\downarrow$ (number of pairs) and $u = N_\uparrow - N_\downarrow$ (number of unpaired majority spins).
• Hamiltonian: The system is modeled using a 2D periodic box with a modified Pöschl-Teller potential to simulate short-range attractive interactions. The interaction strength is tuned via the parameter $v_0$, characterized by the dimensionless parameter $\eta = \ln(k_F a_s)$, where $a_s$ is the 2D scattering length.
• Simulation Details:
  • System Sizes: Simulations were performed on two system sizes: $(N_\uparrow, N_\downarrow) = (13, 5)$ and $(25, 9)$.
  • Optimization: Network parameters were optimized using the variational principle with the Kronecker-factored Approximate Curvature (KFAC) algorithm.
  • Observables: Key quantities computed include particle density, superfluid order parameters, momentum distributions ($n_q$), and the momentum-resolved condensate fraction ($f_q^{\uparrow\downarrow}$).

3. Key Contributions and Results

The study identifies three distinct regimes based on interaction strength, revealing new physics in the intermediate crossover region.

Regime I: Weakly Interacting (BCS Limit, $\eta \gtrsim 0$)

• Observation: The system exhibits the FFLO phase.
• Evidence: The momentum-resolved condensate fraction $f_q^{\uparrow\downarrow}$ shows four distinct peaks at finite momentum vectors ($|q_x| = G, |q_y| = G$), corresponding to the symmetry of the square simulation box.
• Interpretation: Cooper pairs carry a finite center-of-mass momentum to accommodate the spin imbalance, consistent with theoretical predictions for lattice models and diffusion Monte Carlo results.

Regime II: Intermediate Interaction (Crossover, $\eta \approx 0$ to $-1$)

• Discovery: The authors report the observation of a previously unknown crystalline phase.
• Phenomenon:
  • Translational Symmetry Breaking: Unlike the uniform fluid seen in other regimes, the real-space particle density reveals a periodic structure.
  • Structure: The minority-spin fermions (and their paired majority partners) form a crystal lattice (specifically a triangular lattice distorted by the periodic boundary), while the excess unpaired majority-spin fermions form a uniform "sea" surrounding the crystal.
  • Mechanism: This is attributed to a fine balance between the bare attractive interaction (forming tight pairs) and an effective repulsion between these tightly bound bosonic dimers, mediated by the unpaired fermions. This repulsion prevents the collapse into a liquid and stabilizes the crystal.
• Significance: This is the first observation of a $T=0$ crystalline phase in a 2D SIFG with purely attractive interactions, challenging the notion that crystallization (like Wigner crystals) requires repulsive interactions.

Regime III: Strongly Interacting (BEC Limit, $\eta \lesssim -1$)

• Observation: Phase Separation occurs.
• Structure: The system separates into regions of tightly bound bosonic pairs (dimers) and regions containing only unpaired majority-spin fermions.
• Momentum Density:
  • The minority-spin momentum density is depleted almost to zero within the Fermi circle of the unpaired majority spins.
  • This "hole" arises because Pauli exclusion prevents the paired majority spins from occupying states already filled by the unpaired majority spins.
  • The unpaired majority spins behave like a nearly non-interacting Fermi liquid, while the pairs contribute a diffuse momentum halo.
• Consistency: These results align with mean-field BCS calculations in the thermodynamic limit for strong coupling.

4. Significance and Impact

• Methodological Advancement: The paper demonstrates the power of AGPs FermiNet in solving complex, spin-imbalanced fermionic problems where traditional QMC methods struggle. It validates neural quantum states as a robust tool for exploring the BCS-BEC crossover.
• New Physics: The discovery of the crystalline phase in the intermediate regime provides a new perspective on the phase diagram of 2D spin-imbalanced gases. It suggests that effective repulsions between composite bosons can emerge even in purely attractive systems, leading to exotic ordering.
• Experimental Relevance: The findings offer specific predictions (e.g., momentum density holes, specific crystalline structures) that can be tested in future experiments with ultracold atomic gases in optical lattices or 2D traps, where interaction strength and population imbalance are highly tunable.

In summary, this work bridges the gap between theoretical predictions and numerical reality for 2D spin-imbalanced Fermi gases, uncovering a rich phase diagram that includes FFLO states, phase-separated mixtures, and a novel crystalline state of Cooper pairs.