Practical reach of the imaginary-chemical-potential canonical (Fourier-transform) approach to finite-density lattice fermion systems

Simulating interacting fermions at finite density is blocked by the sign problem: at real chemical potential the fermion determinant is complex, so importance-sampled Monte Carlo fails. Alford, Kapustin, and Wilczek (1999) proposed simulating at imaginary chemical potential $\mu=i\nu$, where the determinant is real and positive for an even number of degenerate flavors, and extracting the canonical partition functions $Z(N)$ at fixed fermion number $N$ as Fourier coefficients of the imaginary-$\mu$ grand-canonical partition function. They demonstrated the method on a small Hubbard cluster and identified its structural limit: the Fourier-integration error grows exponentially with $N$, so only small $N$ is reachable and the method fails in the thermodynamic limit. The open problem is to determine the practical reach of this canonical reconstruction — how large a particle number it can deliver at given temperature, volume, Monte Carlo statistics, and numerical precision, and whether the low-temperature regime (where the $N=0$ contribution dominates) is accessible — and, within that reach, the finite-density physics the method was proposed to deliver: pairing and the metal-insulator transition in the doped two-dimensional Hubbard model, and the finite-density QCD phase diagram including the critical region. Later literature diagnosed the instability as catastrophic digit cancellation and mitigated it with multiple-precision Fourier transforms, and canonical lattice-QCD results now reach physical quark masses and $\mu_B\approx500$ MeV; the low-temperature high-density regime, a systematic quantification of the maximal reachable particle number, and the Hubbard application beyond the small-cluster feasibility study remain open.

Background

Finite-density Monte Carlo and the sign problem. In Euclidean lattice field theory, fermions enter the path-integral weight through the determinant of the fermion matrix $M(\mu)$, which depends on the chemical potential $\mu$ — the thermodynamic variable conjugate to a conserved charge, such as baryon number or particle number. At $\mu=0$ this determinant is real and non-negative, so Markov-chain Monte Carlo with importance sampling works; at real $\mu\ne0$ it is in general complex, so the partition function becomes a sum of terms with fluctuating phases whose cancellation error grows exponentially with volume — the sign problem. In the worst case such computations are NP-hard (Troyer and Wiese, 2005), so practical progress relies on problem-specific circumventions rather than a general solution.

The imaginary-chemical-potential canonical method. For an even number of degenerate fermion flavors the determinant at imaginary chemical potential $\mu=i\nu$ is real and positive, so standard algorithms apply. The grand-canonical partition function then has the Fourier expansion $Z(i\nu)=\sum_N Z(N)\,e^{i\beta\nu N}$, where $\beta$ is the inverse temperature and $Z(N)$ is the canonical partition function at fixed fermion number $N$: each $Z(N)$ is a Fourier coefficient of the simulation data. At real chemical potential the canonical functions recombine as $Z(\mu)=\sum_N Z(N)\,e^{\beta\mu N}$, so this route reorganizes (rather than removes) the sign problem. Alford, Kapustin, and Wilczek (1999) formulated it for lattice QCD and tested the numerics on the Hubbard model — the standard model of correlated electrons (fermions hopping on a lattice with on-site repulsion $U$), whose doped (away from half filling) physics of pairing correlations and the metal-insulator transition is itself sign-problem-blocked — using a $4^2\times10$ cluster at $U=0.1$ and $\beta=1.5$ and reaching fermion numbers $N\le6$. They showed that the integrand defining $Z(N)$ oscillates increasingly rapidly with $N$ so that the numerical-integration error grows exponentially with $N$: the method works only for small $N$ and fails in the thermodynamic limit, and extraction becomes harder at low temperature, where the $N=0$ contribution dominates $Z(i\nu)$. They proposed the method as an alternative to the Glasgow method (a fugacity-expansion reweighting from $\mu=0$ ensembles, whose overlap problems produce unphysical early baryon filling) and as a possible route to the QCD phase diagram, including its conjectured critical point at finite baryon density.

Adjacent strategies and their status. Canonical projection methods that extract fixed-$N$ sectors directly predate the source paper (Hasenfratz and Toussaint, 1992; Liu, 2002; Alexandru et al., 2005). Taylor expansion in $\mu$ about $\mu=0$, reweighting, and analytic continuation from imaginary $\mu$ became the standard finite-density lattice-QCD tools (reviews: Lombardo, 2007; Philipsen, 2007). For QCD, imaginary-$\mu$ data are confined by the Roberge-Weiss periodicity — the partition function is periodic in the imaginary quark chemical potential with period $2\pi T/3$, and the associated Roberge-Weiss endpoints bound any analytic continuation in $\mu$. At finite isospin density (equal and opposite chemical potentials for two flavors) QCD has no sign problem and can be simulated directly (Son and Stephanov, 2001), which isolates finite baryon density as the blocked sector. In the Hubbard model, sign-problem severity peaks near its quantum critical points (Mondaini, Tarat, and Scalettar, 2022) — precisely the regime the source paper proposed to study.

Problem Statement

Determine the practical reach of the canonical (Fourier-transform) approach to finite-density lattice fermion systems simulated at imaginary chemical potential. The core open question identified by Alford, Kapustin, and Wilczek (1999) is how high the fermion number $N$ can be pushed before the numerical integration defining the canonical partition function $Z(N)$ becomes undoable. Concretely:

(1) Reach in particle number. Establish the maximal fermion (baryon) number $N_{\max}$ that the imaginary-$\mu$ canonical reconstruction can deliver, as a function of temperature $T$, volume $V$, Monte Carlo statistics, and numerical precision, including the scaling law of the exponential error growth with $N$ identified by Alford, Kapustin, and Wilczek (1999); and determine whether the failure at large $N$ is only quantitative (precision-limited, mitigable by multiple-precision arithmetic and brute-force statistics) or structural. Operationally, $N_{\max}$ is the largest $N$ for which the canonical weight $Z(N)$ is recovered within a stated accuracy tolerance, at a stated number of configurations and working precision. Part (1) consists of establishing (i) an explicit $N_{\max}$ scaling relation in $T$, $V$, statistics, and precision, evaluated at a specified model and lattice setup; (ii) the error-growth scaling law with $N$; and (iii) a determination of whether the large-$N$ failure is quantitative or structural.

(2) Low-temperature reach. Determine whether canonical amplitudes can be extracted in the low-temperature regime where the $N=0$ term dominates the imaginary-$\mu$ partition function — the regime of nuclear-matter onset and of the conjectured QCD critical region — and if so, at what cost in statistics and precision. This is an independent question, not the low-$T$ limit of (1): in addition to $N_{\max}$ as $T$ is lowered, it asks for a feasibility determination for the $N=0$-dominated regime and a separate statistics-and-precision cost estimate.

(3) Physics within the established reach. The source paper proposed the method as a route to sign-problem-blocked finite-density physics: (a) in the Hubbard model, the pairing correlations of the doped model and the metal-insulator transition near half filling, on systems beyond the $4^2\times10$, $U=0.1$ feasibility study and at lower temperatures and stronger couplings; (b) in QCD, the finite-density phase diagram, including the critical region around the conjectured critical endpoint. Since the method fails in the thermodynamic limit (as the source paper established), these applications are open directions that a reach established under (1)-(2) would make accessible.

References

1. Alford, M., Kapustin, A., and Wilczek, F. (1999). Imaginary chemical potential and finite fermion density on the lattice. Physical Review D, 59, 054502. DOI 10.1103/PhysRevD.59.054502. arXiv:hep-lat/9807039. https://doi.org/10.1103/PhysRevD.59.054502
2. Hasenfratz, A., and Toussaint, D. (1992). Canonical ensembles and nonzero density quantum chromodynamics. Nuclear Physics B, 371, 539-549. DOI 10.1016/0550-3213(92)90247-9. https://doi.org/10.1016/0550-3213(92)90247-9
3. Liu, K.-F. (2002). Finite density algorithm in lattice QCD - a canonical ensemble approach. In: Lattice Statistics and Mathematical Physics, pp. 165-180. DOI 10.1142/9789812776358_0020. https://doi.org/10.1142/9789812776358_0020
4. Alexandru, A., Faber, M., Horváth, I., and Liu, K.-F. (2005). Lattice QCD at finite density via a new canonical approach. Physical Review D, 72, 114513. DOI 10.1103/PhysRevD.72.114513. https://doi.org/10.1103/PhysRevD.72.114513
5. Kratochvila, S., and de Forcrand, P. (2005). The canonical approach to Finite Density QCD. Proceedings of Science, LAT2005, 167. DOI 10.22323/1.020.0167. arXiv:hep-lat/0509143. https://arxiv.org/abs/hep-lat/0509143
6. Oka, S., for the Zn-Collaboration (2015). Exploring finite density QCD phase transition with canonical approach -Power of multiple precision computation-. arXiv:1511.04711. https://arxiv.org/abs/1511.04711
7. Bornykov, V. G., Boyda, D. L., Goy, V. A., Molochkov, A. V., Nakamura, A., Nikolaev, A. A., and Zakharov, V. I. (2017). New approach to canonical partition functions computation in $N_f=2$ lattice QCD at finite baryon density. Physical Review D, 95, 094506. DOI 10.1103/PhysRevD.95.094506. arXiv:1611.04229. https://doi.org/10.1103/PhysRevD.95.094506
8. Adam, A., Borsányi, S., Fodor, Z., Guenther, J. N., Parotto, P., Pásztor, A., Pirelli, L., and Wong, C. H. (2025). Finite density lattice QCD without extrapolations. arXiv:2512.09415. https://arxiv.org/abs/2512.09415
9. Adam, A., Borsányi, S., Fodor, Z., Guenther, J. N., Pirelli, L., Parotto, P., Pásztor, A., and Wong, C. H. (2026). Finite density lattice QCD without extrapolation: Bulk thermodynamics with physical quark masses from the canonical ensemble. Physical Review D. DOI 10.1103/tptk-7718. arXiv:2604.14117. https://doi.org/10.1103/tptk-7718
10. Borsányi, S., and Fodor, Z. (2026). The high density frontier in lattice QCD. European Physical Journal Special Topics. DOI 10.1140/epjs/s11734-026-02518-1. https://doi.org/10.1140/epjs/s11734-026-02518-1
11. Hart, A., Laine, M., and Philipsen, O. (2001). Testing imaginary vs. real chemical potential in finite-temperature QCD. Physics Letters B, 505, 141-148. DOI 10.1016/S0370-2693(01)00355-0. https://doi.org/10.1016/S0370-2693(01)00355-0
12. D'Elia, M., and Lombardo, M. P. (2003). Finite density QCD via an imaginary chemical potential. Physical Review D, 67, 014505. DOI 10.1103/PhysRevD.67.014505. https://doi.org/10.1103/PhysRevD.67.014505
13. de Forcrand, P., and Philipsen, O. (2002). The QCD phase diagram for small densities from imaginary chemical potential. Nuclear Physics B, 642, 290-306. DOI 10.1016/S0550-3213(02)00626-0. arXiv:hep-lat/0205016. https://doi.org/10.1016/S0550-3213(02)00626-0
14. Bonati, C., D'Elia, M., Negro, F., Sanfilippo, F., and Zambello, K. (2018). Curvature of the pseudocritical line in QCD: Taylor expansion matches analytic continuation. Physical Review D, 98, 054510. DOI 10.1103/PhysRevD.98.054510. https://doi.org/10.1103/PhysRevD.98.054510
15. Murgana, F., and Ruggieri, M. (2025). Assessing the reconstruction of the critical line in the QCD phase diagram from imaginary to real chemical potential. arXiv:2505.04569. https://arxiv.org/abs/2505.04569
16. Kashiwa, K., and Kouno, H. (2026). The canonical approach at high temperature revisited. arXiv:2605.19287. https://arxiv.org/abs/2605.19287
17. Ding, H.-T. (2026). Lattice QCD at finite temperature and density. arXiv:2603.16230. https://arxiv.org/abs/2603.16230
18. Lin, F., Šmakov, J., Sørensen, E. S., Kallin, C., and Berlinsky, A. J. (2005). Imaginary-chemical-potential quantum Monte Carlo method for Hubbard molecules. Physical Review E, 71, 047701. DOI 10.1103/PhysRevE.71.047701. arXiv:cond-mat/0412641. https://doi.org/10.1103/PhysRevE.71.047701
19. Carlström, J. (2017). Spin-charge transformation of lattice fermion models: duality approach for diagrammatic simulation of strongly correlated systems. Journal of Physics: Condensed Matter, 29, 385602. DOI 10.1088/1361-648X/aa7e1b. arXiv:1607.06293. https://doi.org/10.1088/1361-648X/aa7e1b
20. Filothodoros, E. G. (2024). Fermionic and bosonic partition functions at imaginary chemical potential as Bloch functions. arXiv:2405.09686. https://arxiv.org/abs/2405.09686
21. Filothodoros, E. G. (2026). Thermal phase structure of the attractive Fermi Hubbard model with imaginary chemical potential. arXiv:2604.18798. https://arxiv.org/abs/2604.18798
22. Filothodoros, E. G. (2026). Bose-Fermi mapping in Hubbard models at imaginary chemical potential and phase-induced fermionization. arXiv:2605.10418. https://arxiv.org/abs/2605.10418
23. Son, D. T., and Stephanov, M. A. (2001). QCD at finite isospin density. Physical Review Letters, 86, 592-595. DOI 10.1103/PhysRevLett.86.592. https://doi.org/10.1103/PhysRevLett.86.592
24. Troyer, M., and Wiese, U.-J. (2005). Computational complexity and fundamental limitations to fermionic quantum Monte Carlo simulations. Physical Review Letters, 94, 170201. DOI 10.1103/PhysRevLett.94.170201. https://doi.org/10.1103/PhysRevLett.94.170201
25. Mondaini, R., Tarat, S., and Scalettar, R. (2022). Quantum critical points and the sign problem. Science, 375, 418-424. DOI 10.1126/science.abg9299. https://doi.org/10.1126/science.abg9299
26. Lombardo, M. P. (2007). Lattice QCD at finite temperature and density. Modern Physics Letters A, 22, 457-471. DOI 10.1142/S0217732307023055. arXiv:hep-lat/0509180. https://doi.org/10.1142/S0217732307023055
27. Philipsen, O. (2007). Lattice QCD at finite temperature and density. European Physical Journal Special Topics, 152, 29-60. DOI 10.1140/epjst/e2007-00376-3. https://doi.org/10.1140/epjst/e2007-00376-3