Config of variPEPS module (`varipeps.config`)

class varipeps.config.Grad_Fixed_Point_Method(*values)

Bases: IntEnum

EIGEN_SOLVER = 3: Use eigen solver method to calculate gradient of CTMRG routine

ITERATIVE = 1: Use iterative method to calculate gradient of CTMRG routine

LINEAR_SOLVER = 2: Use linear solver method to calculate gradient of CTMRG routine

class varipeps.config.Line_Search_Methods(*values)

Bases: IntEnum

ARMIJO = 2: Armijo line search method

HAGERZHANG = 4: Hager-Zhang line search method

SIMPLE = 1: Simple line search method

WOLFE = 3: Wolfe line search method

class varipeps.config.Optimizing_Methods(*values)

Bases: IntEnum

BFGS = 3: BFGS method

CG = 2: Conjugate gradient method

L_BFGS = 4: L-BFGS method

STEEPEST = 1: Steepest gradient descent

class varipeps.config.Projector_Method(*values)

Bases: IntEnum

FISHMAN = 3: Use the Fishman method for projector calculation

FULL = 2: Use full network for projector calculation

HALF = 1: Use only half network for projector calculation

HALF_FISHMAN = 4: Use the Fishman method but with half projectors as basis

class varipeps.config.Slurm_Restart_Mode(*values)

Bases: IntEnum

AUTOMATIC_RESTART = 4: Write restart script and start new slurm job with it

DISABLED = 1: Disable automatic restart of slurm job if maximal runtime limit is reached

WRITE_NEED_RESTART_FILE = 2: Write file to indicate that restart is needed but no slurm scripts

WRITE_RESTART_SCRIPT = 3: Write slurm restart script but do not submit new slurm job

class varipeps.config.VariPEPS_Config(ad_use_custom_vjp: bool = True, ad_custom_print_steps: bool = False, ad_custom_verbose_output: bool = False, ad_custom_convergence_eps: float = 1e-07, ad_custom_max_steps: int = 75, ad_custom_fixed_point_method: Grad_Fixed_Point_Method = Grad_Fixed_Point_Method.LINEAR_SOLVER, checkpointing_ncon: bool = False, checkpointing_projectors: bool = False, ctmrg_convergence_eps: float = 1e-08, ctmrg_enforce_elementwise_convergence: bool = True, ctmrg_max_steps: int = 75, ctmrg_print_steps: bool = False, ctmrg_verbose_output: bool = False, ctmrg_truncation_eps: float = 1e-12, ctmrg_fail_if_not_converged: bool = True, ctmrg_full_projector_method: Projector_Method = Projector_Method.FISHMAN, ctmrg_increase_truncation_eps: bool = True, ctmrg_increase_truncation_eps_factor: float = 100.0, ctmrg_increase_truncation_eps_max_value: float = 1e-06, ctmrg_heuristic_increase_chi: bool = True, ctmrg_heuristic_increase_chi_threshold: float = 1e-06, ctmrg_heuristic_increase_chi_step_size: int = 2, ctmrg_heuristic_decrease_chi: bool = True, ctmrg_heuristic_decrease_chi_step_size: int = 1, triangular_ctmrg_use_split: bool = False, svd_sign_fix_eps: float = 0.1, svd_ad_use_lorentz_broadening: bool = False, svd_ad_lorentz_broadening_eps: float = 1e-13, optimizer_method: Optimizing_Methods = Optimizing_Methods.L_BFGS, optimizer_max_steps: int = 300, optimizer_convergence_eps: float = 1e-05, optimizer_ctmrg_preconverged_eps: float = 1e-05, optimizer_fail_if_no_step_size_found: bool = False, optimizer_l_bfgs_maxlen: int = 15, optimizer_preconverge_with_half_projectors: bool = False, optimizer_preconverge_with_half_projectors_eps: float = 0.001, optimizer_autosave_step_count: int = 2, optimizer_random_noise_eps: float = 0.0001, optimizer_random_noise_max_retries: int = 5, optimizer_random_noise_relative_amplitude: float = 0.1, optimizer_reuse_env_eps: float = 0.001, optimizer_use_preconditioning: bool = True, optimizer_precond_gmres_krylov_subspace_size: int = 30, optimizer_precond_gmres_maxiter: int = 3, line_search_method: Line_Search_Methods = Line_Search_Methods.HAGERZHANG, line_search_initial_step_size: float = 1.0, line_search_reduction_factor: float = 0.5, line_search_max_steps: int = 40, line_search_armijo_const: float = 0.0001, line_search_wolfe_const: float = 0.9, line_search_use_last_step_size: bool = False, line_search_hager_zhang_quad_step: bool = True, line_search_hager_zhang_delta: float = 0.1, line_search_hager_zhang_sigma: float = 0.9, line_search_hager_zhang_psi_0: float = 0.01, line_search_hager_zhang_psi_1: float = 0.1, line_search_hager_zhang_psi_2: float = 2.0, line_search_hager_zhang_eps: float = 1e-06, line_search_hager_zhang_theta: float = 0.5, line_search_hager_zhang_gamma: float = 0.66, line_search_hager_zhang_rho: float = 5, line_search_hager_zhang_eps_use_grad_norm: bool = True, line_search_hager_zhang_eps_grad_norm_factor: float = 0.01, basinhopping_niter: int = 20, basinhopping_T: float = 0.001, basinhopping_niter_success: int = 5, spiral_wavevector_type: Wavevector_Type = Wavevector_Type.TWO_PI_POSITIVE_ONLY, slurm_restart_mode: Slurm_Restart_Mode = Slurm_Restart_Mode.WRITE_NEED_RESTART_FILE, jax_compilation_cache_memory_factor: float = 0.5)

Bases: object

Config class for varipeps module. Normally only the blow created instance config is used.

Parameters:

ad_use_custom_vjp (bool, optional) – Use custom VJP rule for the CTMRG routine during AD calculation.
Default: True
ad_custom_print_steps (bool, optional) – Print steps of fix-point iteration in custom VJP function.
Default: False
ad_custom_verbose_output (bool, optional) – Print verbose output in custom VJP function.
Default: False
ad_custom_convergence_eps (float, optional) – Convergence criterion for the custom VJP function.
Default: 1e-07
ad_custom_max_steps (int, optional) – Maximal number of steps for fix-pointer iteration of the custom VJP function.
Default: 75
ad_custom_fixed_point_method (Grad_Fixed_Point_Method, optional) – Select method how the gradient of the CTMRG fixed point routine is calculated.
Default: Grad_Fixed_Point_Method.LINEAR_SOLVER
checkpointing_ncon (bool, optional) – Enable AD checkpointing for the ncon calls.
Default: False
checkpointing_projectors (bool, optional) – Enable AD checkpointing for the the calculation of the proejctors.
Default: False
ctmrg_convergence_eps (float, optional) – Convergence criterion for the CTMRG routine.
Default: 1e-08
ctmrg_enforce_elementwise_convergence (bool, optional) – Enforce elementwise convergence of the CTM tensors instead of only convergence of the singular values of the corners.
Default: True
ctmrg_max_steps (int, optional) – Maximal number of steps for fix-pointer iteration of the CTMRG routine.
Default: 75
ctmrg_print_steps (bool, optional) – Print steps of fix-point iteration in CTMRG routine.
Default: False
ctmrg_verbose_output (bool, optional) – Print verbose output in CTMRG routine.
Default: False
ctmrg_truncation_eps (float, optional) – Value for cut off of the singular values compared to the biggest one. Used in the calculation of the CTMRG projectors.
Default: 1e-12
ctmrg_fail_if_not_converged (bool, optional) – Flag if the CTMRG routine should fail with an error if no convergence can be reached within the maximal number of steps. If disabled, the result converged so far is returned.
Default: True
ctmrg_full_projector_method (Projector_Method, optional) – Set which projector method should be used as default (full) projector method during the CTMRG routine. Sensible values are FULL or FISHMAN.
Default: Projector_Method.FISHMAN
ctmrg_increase_truncation_eps (bool, optional) – Flag if the CTMRG routine should try higher truncation thresholds for the SVD based projector methods if the routine does not converge in the maximum number of steps.
Default: True
ctmrg_increase_truncation_eps_factor (float, optional) – Factor by which the truncation threshold should be increased.
Default: 100.0
ctmrg_increase_truncation_eps_max_value (float, optional) – Maximal value for the truncation threshold. Do not increase higher than this value.
Default: 1e-06
ctmrg_heuristic_increase_chi (bool, optional) – Flag if the CTMRG routine should try higher environment bond dimension for if the routine found singular values above a threshold during the projector calculation of the last absorption step.
Default: True
ctmrg_heuristic_increase_chi_threshold (float, optional) – Threshold for the heuristic environment bond dimension increase.
Default: 1e-06
ctmrg_heuristic_increase_chi_step_size (int, optional) – Step size for the heuristic environment bond dimension increase.
Default: 2
ctmrg_heuristic_decrease_chi (bool, optional) – Flag if the CTMRG routine should try lower environment bond dimension for if the routine found singular values below the SVD threshold during the projector calculation of the last absorption step.
Default: True
ctmrg_heuristic_decrease_chi_step_size (int, optional) – Step size for the heuristic environment bond dimension decrease.
Default: 1
triangular_ctmrg_use_split (bool, optional) – Flag if the split projector method should be used in the triangular CTMRG.
Default: False
svd_sign_fix_eps (float, optional) – Value for numerical stability threshold in sign-fixed SVD.
Default: 0.1
svd_ad_use_lorentz_broadening (bool, optional) – Enable Lorentz broadening in the AD rule for the SVD.
Default: False
svd_ad_lorentz_broadening_eps (float, optional) – Numerical stabilization constant in the Lorentz broadening in the AD rule for the SVD.
Default: 1e-13
optimizer_method (Optimizing_Methods, optional) – Method used for variational optimization of the PEPS network.
Default: Optimizing_Methods.L_BFGS
optimizer_max_steps (int, optional) – Maximal number of steps for fix-pointer iteration in optimization routine.
Default: 300
optimizer_convergence_eps (float, optional) – Convergence criterion for the optimization routine.
Default: 1e-05
optimizer_ctmrg_preconverged_eps (float, optional) – Convergence criterion for the optimization routine using the gradient calculations with the preconverged environment.
Default: 1e-05
optimizer_fail_if_no_step_size_found (bool, optional) – Flag if the optimizer routine should fail with an error if no step size can be found before the gradient norm is below the convergence threshold. If disabled, the result converged so far is returned.
Default: False
optimizer_l_bfgs_maxlen (int, optional) – Maximal number of previous steps used for the L-BFGS method.
Default: 15
optimizer_preconverge_with_half_projectors (bool, optional) – Flag if the optimizer should use only CTM half projectors for the steps till some converge is reached.
Default: False
optimizer_preconverge_with_half_projectors_eps (float, optional) – Convergence criterion for the preconvergence with only the half CTM projectors.
Default: 0.001
optimizer_autosave_step_count (int, optional) – Step count after which the optimizer result is automatically saved.
Default: 2
optimizer_random_noise_eps (float, optional) – Optimizer should try best state sofar with some random noise if gradient norm is below this threshold.
Default: 0.0001
optimizer_random_noise_max_retries (int, optional) – Maximal retries for optimization with random noise.
Default: 5
optimizer_random_noise_relative_amplitude (float, optional) – Relative amplitude used for random noise.
Default: 0.1
optimizer_reuse_env_eps (float, optional) – Reuse CTMRG environment of previous step if norm of gradient is below this threshold.
Default: 0.001
optimizer_use_preconditioning (bool, optional) – Use (local) preconditioning method as described in https://arxiv.org/abs/2511.09546.
Default: True
optimizer_precond_gmres_krylov_subspace_size (int, optional) – Size of Krylov subspace built up during GMRES method for the inversion of the preconditioner.
Default: 30
optimizer_precond_gmres_maxiter (int, optional) – Maximal number of outer iterations inside the GMRES method for the inversion of the preconditioner.
Default: 3
line_search_method (Line_Search_Methods, optional) – Method used for the line search routine.
Default: Line_Search_Methods.HAGERZHANG
line_search_initial_step_size (float, optional) – Initial step size for the line search routine.
Default: 1.0
line_search_reduction_factor (float, optional) – Reduction factor between two line search steps.
Default: 0.5
line_search_max_steps (int, optional) – Maximal number of steps in the line search routine.
Default: 40
line_search_armijo_const (float, optional) – Constant used in Armijo line search method.
Default: 0.0001
line_search_wolfe_const (float, optional) – Constant used in Wolfe line search method.
Default: 0.9
line_search_use_last_step_size (bool, optional) – Flag if the line search should start from the step size of the previous optimizer step.
Default: False
line_search_hager_zhang_quad_step (bool, optional) – Use QuadStep method in Hager-Zhang line search to find initial step size.
Default: True
line_search_hager_zhang_delta (float, optional) – Constant used in Hager-Zhang line search method.
Default: 0.1
line_search_hager_zhang_sigma (float, optional) – Constant used in Hager-Zhang line search method.
Default: 0.9
line_search_hager_zhang_psi_0 (float, optional) – Constant used in Hager-Zhang line search method.
Default: 0.01
line_search_hager_zhang_psi_1 (float, optional) – Constant used in Hager-Zhang line search method.
Default: 0.1
line_search_hager_zhang_psi_2 (float, optional) – Constant used in Hager-Zhang line search method.
Default: 2.0
line_search_hager_zhang_eps (float, optional) – Constant used in Hager-Zhang line search method.
Default: 1e-06
line_search_hager_zhang_theta (float, optional) – Constant used in Hager-Zhang line search method.
Default: 0.5
line_search_hager_zhang_gamma (float, optional) – Constant used in Hager-Zhang line search method.
Default: 0.66
line_search_hager_zhang_rho (float, optional) – Constant used in Hager-Zhang line search method.
Default: 5
line_search_hager_zhang_eps_use_grad_norm (bool, optional) – Use norm of gradient multiplied by VariPEPS_Config.line_search_hager_zhang_eps_grad_norm_factor to calculate eps value in Hager-Zhang line search. If disabled, the fixed value from config parameter VariPEPS_Config.line_search_hager_zhang_eps is used.
Default: True
line_search_hager_zhang_eps_grad_norm_factor (float, optional) – Factor used for gradient based eps calculation. See parameter VariPEPS_Config.line_search_hager_zhang_eps_use_grad_norm for details.
Default: 0.01
basinhopping_niter (int, optional) – Value for parameter niter of scipy.optimize.basinhopping. See this function for details.
Default: 20
basinhopping_T (int, optional) – Value for parameter T of scipy.optimize.basinhopping. See this function for details.
Default: 0.001
basinhopping_niter_success (int, optional) – Value for parameter niterniter_success of scipy.optimize.basinhopping. See this function for details.
Default: 5
spiral_wavevector_type (Wavevector_Type, optional) – Type of wavevector to be used (only positive/symmetric interval/…).
Default: Wavevector_Type.TWO_PI_POSITIVE_ONLY
slurm_restart_mode (Slurm_Restart_Mode, optional) – Mode of operation to restart slurm job if maximal runtime is reached.
Default: Slurm_Restart_Mode.WRITE_NEED_RESTART_FILE
jax_compilation_cache_memory_factor (float, optional) – Limit the jax compilation cache to maximal this factor times the total available memory.
Default: 0.5

class varipeps.config.Wavevector_Type(*values)

Bases: IntEnum

TWO_PI_POSITIVE_ONLY = 1: Use interval [0, 2pi) for q vectors

TWO_PI_SYMMETRIC = 2: Use interval [-2pi, 2pi) for q vectors

Config of variPEPS module (varipeps.config)

Config of variPEPS module (`varipeps.config`)