$\mathbb{Z}_2$ topological invariant in three-dimensional PT- and PC-symmetric class CI band structures

Technical Summary: $Z_2$ Topological Invariant in Three-Dimensional PT- and PC-Symmetric Class CI Band Structures

Problem Statement
The classification of topological invariants for band structures under crystalline symmetries remains incomplete compared to the well-established Altland–Zirnbauer (AZ) classes defined by internal time-reversal and particle-hole symmetries. Specifically, for three-dimensional (3D) systems defined by parity-time ($PT$) and parity-particle-hole ($PC$) symmetries, the classification group for symmetry class CI is known via K-theory to contain a $\mathbb{Z}_2$ component. However, a general, explicit formula for this specific $\mathbb{Z}_2$ invariant on the three-torus (Brillouin zone) has been missing. While a $\mathbb{Z}_2$ index exists for class CI when the parameter space is a three-sphere (defined via Takagi factorization), the subtleties of the three-torus topology prevent a direct application of that definition. This paper addresses the construction of this missing invariant for generic 3D class CI Hamiltonians.

Methodology and Construction
The author constructs the invariant by leveraging the specific geometric and algebraic properties of class CI Hamiltonians:

1. Symmetry Constraints and Bundle Structure:
   Class CI is defined by $PT$ symmetry ($TH^*_k T^\dagger = H_k$) and $PC$ symmetry ($CH^*_k C^\dagger = -H_k$). These symmetries enforce a chiral structure where the flattened Hamiltonian takes the form $H_k = \text{diag}(q_k, -q_k^\dagger)$ with $q_k \in U(n)$ being a symmetric unitary matrix ($q_k^\top = q_k$).
   The $PT$ symmetry allows for a local Takagi factorization $q_k = Q_k Q_k^\top$, where $Q_k \in U(n)$. The transition functions between local patches belong to the orthogonal group $O(n)$, defining a principal $O(n)$ bundle $P_q$ over the Brillouin zone $T^3$. This bundle is equipped with a real (anti-symmetric) Berry connection $A$.

2. Spin-Chern–Simons Action:
   The core of the construction is the definition of a "spin-Chern–Simons" (spin-CS) action. Unlike the standard Chern–Simons action which is defined modulo $2\pi$, the spin-CS action is defined via the $\eta$-invariant of the Dirac operator coupled to the $O(n)$ connection.
   $$\text{CS}_{\text{spin}}(P, A, \sigma) := 2\pi \left( \xi(P, A, g, \sigma) - n\xi_{\text{triv}}(g, \sigma) \right) \mod 4\pi$$
   Here, $\xi$ is the reduced $\eta$-invariant, and $\sigma$ denotes a choice of spin structure on the manifold. The $PT$ symmetry ensures the action has a $4\pi$ periodicity rather than the standard $2\pi$.

3. Quantization via PC Symmetry:
   The $PC$ symmetry imposes a further constraint that quantizes the spin-CS action to the set $\{0, 2\pi\}$. Consequently, the quantity $\nu[q, \sigma] = \frac{1}{2\pi} \text{CS}_{\text{spin}}(P_q, A_q, \sigma) \mod 2$ yields a well-defined $\mathbb{Z}_2$ invariant (values 0 or 1).

4. Spin Structure Dependence:
   The invariant $\nu[q, \sigma]$ explicitly depends on the choice of spin structure $\sigma$ on the torus. The paper analyzes this dependence, showing that while the value of $\nu$ shifts with $\sigma$, the distinction between trivial and non-trivial phases (i.e., whether a system can be adiabatically connected to a trivial one) remains invariant. For the specific case of the three-torus $T^3$, the invariant is quantized to $\{0, 1\}$ regardless of the spin structure, provided the first Stiefel–Whitney class $w_1(P_q)$ satisfies certain conditions (which is always true for $T^3$).

Key Results

• Existence of the Invariant: The paper provides the first explicit construction of the $\mathbb{Z}_2$ invariant for 3D class CI systems on a torus.
• Additivity: The invariant satisfies a sum rule under direct sums of band structures: $\nu[q \oplus q', \sigma] = \nu[q, \sigma] + \nu[q', \sigma] \mod 2$.
• Consistency with Known Cases: In scenarios where a global Takagi factorization exists (i.e., $q_k = Q_k Q_k^\top$ globally), the new invariant reduces to the parity of the 3D winding number of $Q_k$, consistent with previous results for the three-sphere.
• Distinguishing Power: The paper demonstrates through explicit lattice models that $\nu$ can distinguish topological phases that share identical values for all previously known invariants (winding numbers $W_1$ and second Stiefel–Whitney classes $w_2$).
  • Example: Two models, $q_A$ and $q_B$, are constructed with identical trivial $W_1$ and identical non-trivial $w_2$ classes. However, $\nu[q_A, \sigma_{NS}] = 1$ while $\nu[q_B, \sigma_{NS}] = 0$, proving they are topologically distinct and cannot be connected adiabatically.
• Spin Structure Nuance: The paper clarifies that while the invariant depends on the spin structure, this dependence does not introduce physical inconsistencies. The trivial phase is uniquely represented by the constant matrix $q=1_n$, which is independent of spin structure, ensuring the classification of phases remains robust.

Significance and Claims
The paper claims to complete the classification of topological invariants for the eight real AZ classes defined by $PT$ and $PC$ symmetries in spatial dimensions up to three. By constructing this previously missing $\mathbb{Z}_2$ invariant, the author resolves a gap in the theoretical framework for topological band theory in crystalline systems. The work emphasizes that the invariant is a necessary tool for distinguishing phases in class CI that are invisible to standard winding numbers or Stiefel–Whitney classes alone. The author notes that the paper does not address specific physical phenomena like corner states (which were discussed in prior literature) but focuses strictly on the mathematical construction and classification of the bulk invariant.