spinw.anneal method

performs simulated annealing of spins

Syntax

stat = anneal(obj,Name,Value)

Description

stat = anneal(obj,Name,Value) performs simulated annealing on the spin Hamiltonian defined in obj. It assumes a classical spin system and employs the Metropolis–Hastings algorithm for state updates. The annealing is performed from a given initial temperature down to final temperature with user defined steps and number of Monte-Carlo cycles per temperature. The spinw.anneal can also measure different thermodynamic quantities, such as heat capacity. The function can deal with single ion anisotropy and arbitrary exchange interactions. It can also restrict the spin degrees of freedom from 3 (default) to 2 or 1 to speed up simulation on xy and Ising systems. For these restricted dimensions only isotropic exchange interactions are allowed. Also the g-tensor is assumed to be 2.

Warning: The calculated energies does not contain the self energy (spin coupled to itself), thus the energies calculated here can differ from the output of spinw.energy.

Input Arguments

obj: spinw object.

Name-Value Pair Arguments

'spinDim'

Dimensionality of the spins:

1 Ising spins.
2 XY spins.
3 Heisenberg spins, default.

For Ising (spinDim=1) and XY (spinDim=2) models only isotropic exchange interaction and magnetic field can be used. For Ising the direction of the spins are along x-axis, for XY model the the xy-plane. Magnetic fields perpendicular to these directions are omitted.

'initT'

The initial temperature, can be any positive number in units of Kelvin. Default value is 1.

'endT'

Temperature at which the annealing will stop, can be any positive number smaller than initT, unit is Kelvin. Default value is $10^{-3}$ .

'cool'

Defines how the following temperature value is calculated from the previous one using a user function. Any function that takes a scalar as input and returns a smaller but positive scalar as output. Default is @(T)(0.92*T).

'random'

If true the initial spin orientation will be random, which is effectively a $T=\infty$ paramagnet. If the initial spin configuration is undefined (obj.magstr.S is empty) the initial configuration is always random independently of this parameter. Default value is false.

'nMC'

Number of Monte-Carlo steps per spin at each temperature step which includes thermalization and the sampling for the extracted TD properties at the last temperature). Default is 100.

'nORel'

Number of over-relaxation steps after every Monte-Carlo steps. It rotates the spins around the direction of the local field by 180°. It is reversible and microcanonical if the single ion anisotropy is 0. Default value is 0, to omit over-relaxation.

'nStat'

Number of cycles at the last temperature to calculate statistical averages. It has to be smaller or equal $n_{MC}$ . Default value is 100.

'boundary'

Boundary conditions of the extended unit cell:

free Free, interactions between extedned unit cells are omitted.
per Periodic, interactions between extended unit cells are retained. Default value is {'per' 'per' 'per'}.

'verbosity'

Controls the amount of output to the Command Window:

0 Suppresses all output.
1 Gives final report only.
2 Plots temperature changes and final report, default value.

'nExt'

The size of the magnetic cell in number of unit cells that can override the value stored in obj.magstr.N_ext, given by a row vector with three integers

'fStat'

Function handle to measure TD quantities at the final temperature for nStat Monte-Carlo steps. The function returns a single structure and takes fixed input parameters:

parOut = fStat(state, parIn, T, E, M, nExt).

The function is called once before the annealing process when state=1 to initialise the parameters. The function is called after every Monte-Carlo cycle with state=2 and the output of the previous function call is assigned to the input struct. fStat is called once again in the end with state=3 to calculate final parameters (in the last run, input parIn.param contains all the annealing parameters). For comparison see the defaul function sw_fstat. Default value is @sw_fstat.

'fSub'

Function to define sublattices for Monte-Carlo speedup. Function handle with the following header:

cGraph = fSub(conn,nExt)

where cGraph is a row vector with $n_{magExt}$ number of elements conn is a matrix with dimensions of $[2\times n_{conn}]$ $n_{ext}$ is equal to nExt. For the SpinW implementation see sw_fsub. Default value is @sw_fsub.

'subLat'

Vector that assigns all magnetic moments into non-interacting sublattices, contains a single index $(1,2,3...)$ for every magnetic moment, row vector with $n_{magExt}$ number of elements. If undefined, the function defined in fSub parameter will be used to partition the lattice.

'title'

Gives a title string to the simulation that is saved in the output.

'autoK'

Bin length of the autocorrelation vector. Should be a few times smaller than nMC. Default value is 0 when no autocorrelation function is calculated.

Output Arguments

stat struct that contains the calculated TD averages and the parameters of the simulation with the following fields:

param: All input parameter values of the anneal function.
obj: The clone of the input obj updated with the final magnetic structure.
M: Components of the magnetisation after the last annealing run stored in a matrix with dimensions of $[3\times n_{magExt}]$ .
E: Energy of the system after the last annealing run, excluding the self energy.
T: Final temperature of the sample.

Depending on the fStat parameter, additional fields are included. Using the default function sw_fstat the following parameters are also calculated:

avgM: Average value of the magnetisation vector sampled over nStat runs, stored in a matrix with dimensions of $[3\times n_{magExt}]$ .
stdM: Standard deviation of the mgnetisation vector sampled over nStat runs, stored in a matrix with dimensions of $[3\times n_{magExt}]$ .
avgE: Average system energy per spin averaged over nStat runs, scalar.
stdE: Standard deviation of the system energy per spin over nStat runs, scalar.
Cp: Heat capacity of the sample, calculated using the formula $(\langle E^2\rangle-\langle E\rangle^2)/k_B/T^2$ .
Chi: Magnetic susceptibility of the sample calculated using the formula $(\langle M^2\rangle-\langle M\rangle^2)/k_B/T$ .

Reference

Kirkpatrick, S., Gelatt, C.D., & Vecchi, M.P. (1983). Optimization by Simulated Annealing. Science, 220, 671-680.