Dynamical Mean Field Theory and Impurity Solvers

Dynamical Mean Field Theory and Impurity Solvers

List of Parameters

Physical parameters

NameDescription
Uthe Hubbard interaction U
BETAthe inverse temperature
MUthe chemical potential
Hthe magnetic field in the quantization axis (conventionally $z$) direction (BUT: the solvers do ignore the variable!)
SITESnumber of impurity sites (for DMFT: 1)
FLAVORSnumber of flavors/orbitals of the impurity (commonly 2: spin up/down)
tin case of Bethe lattice it does provide the hopping (the bandwidth is then $W=4t$, the half-bandwidth is $D=2t$); if the option TWODBS is switched on then it does set the nearest-neighbor hopping on the square or hexagonal lattice
t0, t1, …(available currently only for selfconsistency loop in imaginary time) sets the hopping for the Bethe lattice in multiband case (flavors 2i and 2i+1 share the same parameter ti)
Jcoupling for the multiband problems
U'(by default U-2J)
tprimeapplies only if the option TWODBS is switched on and only for the square lattice, then it does set the next-nearest-neighbor hopping
TWODBS(by default sets the square lattice) you may choose either square or hexagonal lattice

Parameters for the self-consistency loop

NameDescription
OMEGA_LOOPset it 1 unless you want to work with semicircular density of states (corresponding to the Bethe lattice in infinitely many dimensions)
ANTIFERROMAGNETif 1 then the antiferromagnetic self-consistency loop will be employed (formula 97 in review ‘96 of A.Georges et al)
SYMMETRIZATIONif 1 then paramagnetic solution is enforced (in versions before 2.1: there has been a misspelling SYMMATRIZATION at several places and a usage of both, SYMMETRIZATION and set to the same value was required)
MAX_ITmaximum number of iteration in self-consistency loop (usually 10-20 will be enough)
CONVERGEDcriterium for stopping the self-consistency loop before reaching MAX_IT - if the maximum change in Green’s function in Matsubara representation is less than CONVERGED, the loop will stop
TOLERANCE(only for hirschfyesim) as above
RELAX_RATE(by default 1; currently implemented only for selfconsistency loop with OMEGA_LOOP switched on) the new Green’s function are in general computed as RELAX_RATE * $G_{new}(i\omega_n)$ + (1-RELAX_RATE) * $G_{old}(i\omega_n)$, which may help if oscillations occur

General parameters

NameDescription
GENERAL_FOURIER_TRANFORMERset it on if you have OMEGA_LOOP and other than the Bethe lattice
EPS_i (i=0,1,…,FLAVORS-1)potential shift for the flavor i (necessary for GENERAL_FOURIER_TRANSFORMER)
EPSSQ_i (i=0,1,…,FLAVORS-1)the second moment of the bandstructure for the flavor i (necessary for GENERAL_FOURIER_TRANSFORMER)
DOSFILEsets the name for the file containing the density of states (expected 2 columns with energy value and corresponding density of states at that energy; equidistant energies required; odd number of rows required due to Simpson integration)
TWODBSswitches on the Hilbert transformation for 2-dimensional systems, currently supported square lattice (with nearest and next-nearest neighbor hoppings) and hexagonal lattice (with nearest neighbor hoppings) [Note: a different 2-dimensional lattice may be easily added]
Loptional parameter available in case of TWODBS is on; defines the half of the linear discretization in the integration in the self-consistency (default: 200)
SOLVERspecifies the impurity solver (“Hybridization” or “Interaction Expansion”; the solver “Hirsch-Fye” does suffer from discretization errors and is thus not recommended)

Parameters for the initial/final Weiss field

NameDescription
H_INITmagnetic field in the quantization axis (conventionally $z$) direction, which is used in computation of the non-interacting initial G0 (if it is not loaded)
G0OMEGA_INPUTname for the text file specifying the Weiss field in Matsubara frequencies $i\omega_n$ (expected 1+FLAVORS columns, and total NMATSUBARA rows) (use only with OMEGA_LOOP)
G0TAU_INPUTname for the text file specifying the Weiss field in imaginary time representation (expected 1+FLAVORS columns, and total N+1 rows) (only with OMEGA_LOOP switched off) |
| GOMEGA_input | specifies the name for the text file where the initial G0 in Matsubara representation will be written (by default it is not written, as it is identical with G0_omega_1) |
| G0TAU_input | name for the text file for the output of the initial G0 in imaginary time (by default it is not written, as it is identical with G0_tau_1) |
| G0OMEGA_output | name for the output file containing the final Weiss field in Matsubara frequencies (by default G0omega_output)(with OMEGA_LOOP) |
| G0TAU_output | name for the output file containing the final Weiss field in Matsubara frequencies (by default G0tau_output) (with OMEGA_LOOP off) |
| INSULATING | if you have specified this option, then the initial G0 will be set up in the insulating limit |

Parameters setting the precision of representation of the Green’s function and the Weiss field

NameDescription
NMATSUBARAnumber of Matsubara frequencies used to represent the Green’s function and the Weiss field (usually equals N)
Nnumber of bins for the Green’s function and the Weiss field in imaginary time (represented in total by N+1 values) (recommended: roughly 1000 for the continuous-time solvers)

Hybridization1 expansion impurity solver parameters

NameDescription
MAX_TIMEsets the maximum time given in seconds spent on the impurity problem solving (basically this sets the duration of a single iteration)
SWEEPSnumber of desired sweeps performed during the calculation (recommendation: set it very high, e.g. $10^9$ and the solver will stop on the time limit given by MAX_TIME)
THERMALIZATIONnumber of sweeps before the Monte Carlo measurements in order to reach configuration close to equilibrium (of the order of 1000)
EPSSQAVthe second moment of the bandstructure (necessary if you have specified your own DOSFILE)
N_ORDERsetting histogram size (if the hybridization order is larger then it will be not stored in the histogram) (value of the order of 100 might be reasonable)
N_MEASnumber of Monte Carlo steps between measurements (of the order of 10000)
N_SHIFTnumber of shifts of segments in a single Monte Carlo step (apparently unused, so 0)
MEASURE_FOURPOINTif switched on then the four-point correlators are being measured
N4point(only used if MEASURE_FOURPOINT is on) description missing so far
CHECKPOINTfilename prefix for checkpointing files and for the final h5 and xml output

Interaction expansion1 impurity solver parameters

NameDescription
MAX_TIMEsets the maximum time given in seconds spent on the impurity problem solving
SWEEPSnumber of desired sweeps performed during the calculation (recommendation: set it very high, e.g. $10^9$ and the solver will stop on the time limit given by MAX_TIME)
THERMALIZATIONnumber of sweeps before the Monte Carlo measurements in order to reach configuration close to equilibrium (of the order of 1000)
SWEEP_MULTIPLICATOR(default: 1)
NRUNS(default: 1)
ALPHA
RECALC_PERIOD(default: 5000)
MEASUREMENT_PERIOD(default: 200)
CONVERGENCE_CHECK_PERIOD(default provided)
ALMOSTZERO(default: $10^{-16}$)
NSELF(default: 10N)
NMATSUBARA_MEASUREMENTS(default: NMATSUBARA)
HISTOGRAM_MEASUREMENT(default: false)
GET_COMPACTED_MEASUREMENTS
ATOMIC
TAU_DISCRETIZATION_FOR_EXP
CHECKPOINTfilename prefix for the checkpointing files and for the final h5 and xml output

Additional parameters

NameDescription
SEEDrandom seed for the pseudorandom generator
RNGpseudorandom generator used (default is “mt19937”), might be switched to “lagged_fibonacci607”

Usage notes

  • Remark on bipartite lattices: the ANTIFERROMAGNET option does assume a Neel-like ordering and requires thus a bipartite lattice. Note that on a bipartite lattice the density of states is symmetric (unless you apply a global potential shift).
  • Since revision 6217, if you provide the DOSFILE or if you use TWODBS and if none of the parameters EPS_i, EPSSQ_i, EPSSQAV is set, then the EPS_i will be set to the first moment of the normalized DOS (in case of TWODBS: 0) and the EPSSQ_i and EPSSQAV will be set to the second moment of the normalized DOS using the provided density of states (in case of TWODBS: using the hard-coded values).
  • Since revision 6217 you may use TWODBS=“hexagonal” to simulate the 2-dimensional hexagonal lattice (nearest-neighbor hoppings only). If you use TWODBS with other value, square lattice is assumed.

Input/output files

The files with prefix BASENAME: (where BASENAME is the name of the parameter input file)

  • BASENAME: it is the input file to be loaded by the application dmft
  • BASENAME.h5: contains the iteration resolved impurity Green’s function $G(\tau)$ and the Weiss field $G^0(\tau)$ in the imaginary time representation; if the selfconsistency loop has been performed in Matsubara representation (= if OMEGA_LOOP has been on) then there will be stored the $G(i\omega_n)$ and $G^0(i\omega_n)$ as well. The selfenergy is there not stored directly, but may be obtained via Dyson equation easily (look into DMFT-01 An introduction to DMFT)

The output/input files in Matsubara representation: (text file which consists of NMATSUBARA rows, each for one Matsubara frequency)

  • G_omega_i (G0_omega_i): contains the imaginary part of the Green’s function (Weiss field) given in Matsubara frequencies after the i-th iteration; rows contain the $\omega_n$ followed by the imaginary part of the Green’s function (Weiss field) for each flavor; thus there are 1+FLAVORS columns in the file
  • G_omegareal_i (G0_omegareal_i): the same as above for the real part
  • selfenergy_i: contains the selfenergy after the i-th iteration; each row consists of $\omega_n$ followed by the real and imaginary part of the selfenergy for each flavor; thus there are 1+2FLAVORS columns in the file
  • G0omega_output (unless not specified differently by the variable G0OMEGA_output): contains the n (corresponding to $\omega_n=\frac{(2n+1)\pi}{\beta})$ followed by the complex Weiss field for each flavor; thus there is one integer column followed by FLAVORS columns of complex numbers defined by the real and imaginary part in brackets
  • G0OMEGA_INPUT: variable specifying the input file with the initial Weiss field in Matsubara representation; does expect the same format as the above output file; thus you may copy it and start a simulation from it

The output/input files in imaginary time representation: (text file which consists of $N+1$ rows, each for one imaginary time $\in\langle 0,\beta\rangle$)

  • G_tau_i (G0_tau_i): contains the (real) Green’s function (Weiss field) after the i-th iteration; rows contain the $\tau_n$ followed by the Green’s function (Weiss field) for each flavor; thus there are 1+FLAVORS columns in the file
  • G0tau_output (unless not specified differently by the variable G0TAU_output): contains the n (corresponding to $\tau_n=\frac{n}{N}\beta$) followed by the complex Weiss field for each flavor; thus there is one integer column followed by FLAVORS columns of complex numbers defined by the real and imaginary part in brackets; in total $N+1$ rows
  • G0OMEGA_INPUT: variable specifying the input file with the initial Weiss field in imaginary time representation; does expect the same format as the above output file; thus you may copy it and start a simulation from it

The output files with prefix given by the optional variable CHECKPOINT:

  • CHECKPOINT.h5: contains the measurements for each iteration
  • CHECKPOINT.xml: contains the input parameters and run information
  • CHECKPOINT.run*: contains information to rerun the simulation (these are the true checkpoints); for each process

The output files for the hybridization expansion impurity solver: (text files)

  • overlap: i-th row contains the $\langle n_\downarrow n_\uparrow\rangle$ in the i-th iteration
  • matrix_size:

Literature

  • A review on DMFT: A. Georges, G. Kotliar, W. Krauth, and M. J. Rozenberg, Dynamical mean-field theory of strongly correlated fermion systems and the limit of infinite dimensions, Rev. Mod. Phys. 68, 13 (1996).
  • On the hybridization expansion impurity solver: P. Werner and A. J. Millis, Hybridization expansion impurity solver: General formulation and application to Kondo lattice and two-orbital models, Phys. Rev. B 74, 155107 (2006).