Skip to content
qiongzPublic

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

151 Commits

Folders and files

Repository files navigation

BLQH

Exact diagonalization of quantum Hall systems on the torus, in the lowest Landau level. A monolayer, a spinless bilayer, and a spinful bilayer share one pipeline: a flavor list builds a Fock sector, the sector builds one sparse Hamiltonian, and a dense eigensolver or Lanczos diagonalizes it.

The previous three programs are kept on the legacy branch.

Build

Requires C++17, OpenMP, and oneMKL (LP64, GNU OpenMP threading). The same link line is used on Intel and on AMD EPYC. icpc is not required.

cmake -S . -B build -DMKLROOT=/opt/intel/oneapi/mkl/2025.2
cmake --build build -j

That builds build/fqhe, including the MKL Lanczos kernel and the launcher that calls the Gluon kernel. The Gluon kernel itself is Python (python/blqh/lanczos_gluon.py). CMake does not compile it. At run time it needs python3 on PATH (or BLQH_PYTHON), a visible GPU, and a PYTHONPATH that can import Triton. The program prepends python/ from this tree.

The executable is build/fqhe. Regression:

OMP_NUM_THREADS=80 MKL_NUM_THREADS=80 ./build/fqhe check 80
./build/fqhe check 80 full    # also the 6-electron monolayer Laughlin state

The square-torus Laughlin alignment, ground-state energies and the dispersion $E(k\ell)$, is in docs/alignment.md. That is the numerical example a new build should match before any other scan.

check verifies:

  • Monolayer Laughlin $\nu=1/3$: the full spectra for 4 and 5 electrons, and, with full, the 6-electron ground-state energy and its 3-fold degeneracy
  • Bilayer Halperin 111: $N_\phi=8$, 4 electrons per layer, $d=0$, $J=K=4$, ground-state energy per electron $-0.636707$
  • Analytic level shifts from bias, Zeeman, interlayer tunneling, and spin-flip tunneling

Parameters

Describe the two layers with Layers, or pass the same fields on the command line.

Flag Meaning
--nphi Number of flux quanta $N_\phi$
--upper, --lower Electrons in each layer, or a filling p/q (nphi * p must be divisible by q)
--spin Four flavors: upper up, upper down, lower up, lower down. Omit it for two layers and no spin
--free Fix only the total electron number. Required when tunneling changes layer or spin
--d Layer separation $d/\ell$
--bias Interlayer potential. Positive raises each upper-layer electron by $\Delta_V/2$
--zeeman Zeeman energy. Positive raises each spin-up electron by $\Delta_Z/2$
--tunnel Spin-conserving interlayer tunneling
--spin-flip Spin-changing interlayer tunneling
--J, --K Magnetic-translation sector. -1 means no projection

If the four spin counts are omitted, a spinful run puts every electron in spin up. Set them with --upper-up, --upper-down, --lower-up, and --lower-down.

One determinant is a uint64_t. Each flavor occupies a contiguous block of $N_\phi$ bits, so $N_\phi\le 16$ with four flavors and $N_\phi\le 32$ with two. Pauli exclusion is the rule that a bit is set at most once.

Commands

# Upper layer at 1/3, lower layer at 2/3, with spin
./build/fqhe point --nphi 6 --upper 1/3 --lower 2/3 --spin --free \
  --d 1 --bias 0 --zeeman 0.05 --tunnel 0.02 --wf 8

# Lowest energy versus layer separation
./build/fqhe scan-d --nphi 6 --upper 1/2 --lower 1/2 --spin --free \
  --d 0:2:0.25 --zeeman 0.05 --out scan_d.csv

# Bias versus tunneling. Use --y spin-flip to sweep the spin-flip channel
./build/fqhe scan-phase --nphi 6 --upper 1/2 --lower 1/2 --spin --free \
  --x bias --y tunnel --bias -0.2:0.2:0.05 --tunnel 0:0.2:0.05 \
  --d 1 --zeeman 0.05 --out phase.csv

# d/l versus filling imbalance Delta nu = nu_upper - nu_lower. Total filling defaults to 1
./build/fqhe scan-imbalance --nphi 6 --nu-total 1 \
  --d 0:2:0.25 --delta-nu 0:1 --out imbalance.csv
# Export one Laughlin CSR. J < 0 is the full Fock space.
./build/fqhe dump-ham --nel 6 --J -1 --out ham.npz

# Lanczos kernel: scalar (default), mkl, or gluon. Gluon accepts at most 40 steps.
./build/fqhe point --nphi 6 --upper 2 --lower 1 --lanczos --steps 40 \
  --lanczos-kernel mkl --threads 64

Dimensions up to 800 are diagonalized densely, and a few excitations above the ground state are reported. Larger sectors use Lanczos and keep the ground state. --dense or --lanczos forces the choice. --lanczos-kernel mkl or gluon also forces Lanczos. scalar is the loop in src/solve.cpp. mkl uses mkl_sparse_z_mv and the --threads count (the flag defaults to 80). gluon keeps the Krylov basis on the GPU.

Columns and order parameters are defined in docs/usage.md.

Layout

include/blqh/fqhe.hpp    parameters, sector, Hamiltonian, solvers, order parameters
src/sector.cpp           flavors, combinations, magnetic translations
src/hamiltonian.cpp      Coulomb interaction, one-body fields, CSR
src/solve.cpp            dense ZHEEVD and the default scalar Lanczos
src/lanczos_mkl.cpp      optional MKL sparse Lanczos
src/lanczos_gluon.cpp    writes the CSR to the Gluon kernel and reads the ground state back
src/dump.cpp             fqhe dump-ham
src/observe.cpp          order parameters
src/cli.cpp              command line
src/check.cpp            regression
python/blqh/             Gluon CSR matvec and Lanczos
tests/ref/               monolayer reference spectra

The dense path calls MKL LAPACKE_zheevd. Lanczos reads the same CSR and returns the ground state in the Fock basis. check still uses the scalar Lanczos.

License

MIT

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages