Skip to content

Pairing Benchmark Observable Contract

This solver shares observable definitions with the number-conserving solver at src/number_conserving/. Definitions that exist in the no-pairing benchmark must keep the same normalization, symbols, and \(K\)-point convention here.

Shared Triangular Geometry

Both solvers use the triangular primitive vectors

\[ \mathbf a_1=(1,0),\qquad \mathbf a_2=\left(\frac12,\frac{\sqrt3}{2}\right), \]

and reciprocal vectors

\[ \mathbf b_1=2\pi\left(1,-\frac{1}{\sqrt3}\right),\qquad \mathbf b_2=2\pi\left(0,\frac{2}{\sqrt3}\right). \]

Thus \(\mathbf b_i\cdot\mathbf a_j=2\pi\delta_{ij}\). On a \(3\times3\) lattice the benchmark \(K\) point is

\[ K=\frac{2}{3}\mathbf b_1+\frac{1}{3}\mathbf b_2 =\left(\frac{4\pi}{3},0\right), \]

implemented as Latt%inv_cell_list(2*Nlx/3+1, Nly/3+1).

Observables Inherited From The No-Pairing Benchmark

symbol DQMC/ED key definition
\(\rho\) density_total \(\langle N_b+N_c\rangle/N_s\)
\(E\) energy_density paper-facing total physical energy \(E=N_s\,\texttt{energy\_density}=\langle H_{\rm phys}\rangle\), excluding \(-\mu N\); plots use \(-E\)
\(D_{bc}\) doubleOcc \(N_s^{-1}\sum_i\langle n_{b,i}n_{c,i}\rangle\)
\(M_b^{(2)}\) onsite_n2_up \(\sum_i\langle n_{b,i}^2\rangle\)
\(M_c^{(2)}\) onsite_n2_do \(\sum_i\langle n_{c,i}^2\rangle\)
\(N_b^2\) numsquare_up \(\langle(\sum_i n_{b,i})^2\rangle\)
\(N_c^2\) numsquare_do \(\langle(\sum_i n_{c,i})^2\rangle\)
\(S_{\rm SF}(K)\) S_SF_K / sf_K \(N_s^{-2}\sum_{ij,\alpha}e^{iK\cdot(r_i-r_j)}\langle a^\dagger_{\alpha i}a_{\alpha j}\rangle\)
\(S_{\rm DW}(K)\) S_DW_K / dw_K \(N_s^{-2}\langle n_K n_{-K}\rangle\), \(n_i=n_{b,i}+n_{c,i}\)
\(S_{\rm PSF}(\Gamma)\) S_PSF_Gamma / psf_Gamma \(N_s^{-2}\sum_{ij}\langle b_i^\dagger c_i^\dagger c_j b_j\rangle\)
\(\mathrm{IPR}_\rho\) IPR \(\sum_i\rho_i^2/(\sum_i\rho_i)^2\), \(\rho_i=\langle n_{b,i}+n_{c,i}\rangle\)

The \(K\)-point structure factors are Hermitian positive structure-factor channels with exact real non-negative thermal averages. DQMC files remain two-column complex files because the Fourier sums are accumulated as complex numbers. The real part is the comparison channel; the imaginary part is checked with the analysis gate \(|{\rm Im}|\le \max(10^{-14},3\,{\rm SEM}_{\rm Im})\), where the absolute floor prevents roundoff-scale values from being amplified by a numerically tiny SEM. A violation of this imaginary-part gate, or a real part that is negative beyond the same roundoff/block-error scale, marks the observable unreliable.

Pairing-Specific Extensions

The pairing benchmark adds observables that do not exist in the no-pairing model. The operators in this table follow the solver convention b_code=b_paper, c_code=-c_paper, in which the real pair coefficient is positive. Consequently the manuscript's anomalous amplitude per site is \(P_{\rm paper}=N_s^{-1}\sum_i\langle b_i c_i+b_i^+c_i^+\rangle_{\rm paper} =-\texttt{pair_equal}\), and its energy contribution per site is \(e_\Delta=-\Delta P_{\rm paper}=\Delta\,\texttt{pair_equal}\). The four plotted benchmark observables are invariant under this phase change. See the Hamiltonian conventions.

symbol key definition
\(P_\Delta\) pair_equal \(N_s^{-1}\sum_i\langle b_i c_i+b_i^\dagger c_i^\dagger\rangle\)
\(e_\Delta\) pairing_energy_density \(\Delta P_\Delta\), per site; total \(E_\Delta=N_s e_\Delta\)
\(e_U\) interaction_energy_density \((U_1+U_2)(M_b^{(2)}+M_c^{(2)})/N_s+2(U_1-U_2)D_{bc}\), per site
\(e_\mu\) chemical_energy_density \(-\mu\rho\), per-site diagnostic only
\(E_{\rm grand}\) grand_energy_density \(N_s\,\texttt{grand\_energy\_density}=E+N_s e_\mu\), diagnostic only

The paper-facing energy panel remains \(-E\), not \(-e\) and not \(-E_{\rm grand}\). The internal DQMC/ED key energy_density is kept only as a storage and comparison convention.

Extending observables

When changing a shared observable, compare the implementations in src/pairing/src/obser_equal.f90 and src/number_conserving/src/obser_equal.f90 as well as their ED operators. For K-point quantities, check reciprocal coordinates, Hermiticity, and normalization. Add the corresponding small ED identity or limiting-case test.