Boris Computational Spintronics

User manual, version 2.8

Dr. Serban Lepadatu, 13th July 2020

Abstract

This manual describes Boris Computational Spintronics, a multi-physics and multi-

scale research software designed to solve three-dimensional magnetization
dynamics problems, coupled with a self-consistent charge and spin transport solver,
heat flow solver with temperature-dependent material parameters, in arbitrary multi-
layered structures and shapes. The computational routines run both on central
processors and graphics processors using the CUDA platform. In addition to simple
user control, advanced simulation configurations are made possible using Python
scripts. The software is open source and currently runs on Windows 7, Windows 10,
and Linux-based 64-bit operating systems, and was programmed using C++17,
CUDA C, and Python.

boris-spintronics.uk/download
https://github.com/SerbanL/Boris2

1
Disclaimer

Boris Computational Spintronics is a freely available research, design and

educational software. The author assumes no responsibility whatsoever for its use
by other parties, and makes no guarantees, expressed or implied, about its quality,
reliability, or any other characteristic. If using Boris for published research please use
a relevant reference as given in the “Selected Publications using Boris” section (an
article describing Boris, which may be used as a complete reference in the future, is
pending).

2
Contents

Installation - Windows ............................................................................................. 6

Installation – Linux-based OS ................................................................................. 7

Overview ................................................................................................................... 9

Tutorial 0 – Quick-Start .......................................................................................... 10

Tutorial 1 – Introduction ........................................................................................ 24

Tutorial 2 – Data Output......................................................................................... 31

Tutorial 3 – Further Data Output ........................................................................... 39

Tutorial 4 – Domain Walls...................................................................................... 43

Tutorial 5 – Domain Wall Movement and Data Processing ................................. 49

Tutorial 6 – ODE Control and Setting Shapes ..................................................... 53

Tutorial 7 – Magnetocrystalline Anisotropy ......................................................... 57

Tutorial 8 – Anisotropic Magneto-Resistance...................................................... 60

Tutorial 9 – Scripting using Python ...................................................................... 67

Tutorial 10 – Current-Induced Domain Wall Movement ...................................... 69

Tutorial 11 – Oersted Fields .................................................................................. 72

Tutorial 12 – Surface Exchange, Multi-Layered Demagnetization and CUDA ... 76

Tutorial 13 – Dzyaloshinskii-Moriya Exchange .................................................... 82

Tutorial 14 – Simulations with non-zero temperature ......................................... 85

3
Tutorial 15 – Thermal Fields .................................................................................. 88

Tutorial 16 – Heat Flow Solver and Joule Heating............................................... 89

Tutorial 17 – Spin Transport Solver ...................................................................... 94

Tutorial 18 – Spin Hall Effect ................................................................................. 99

Tutorial 19 – Spin Pumping and Inverse Spin Hall Effect ................................. 102

Tutorial 20 – Ferromagnetic Resonance ............................................................ 105

Tutorial 21 – Ferromagnetic Resonance with Spin Torques ............................ 111

Tutorial 22 – CPP-GMR ........................................................................................ 115

Tutorial 23 – Skyrmion Movement with Spin Currents ..................................... 118

Tutorial 24 – Roughness and Staircase Corrections ........................................ 123

Tutorial 25 – Defects and Impurities ................................................................... 130

Tutorial 26 – Polycrystalline and Granular Films .............................................. 133

Tutorial 27 – Periodic Boundary Conditions ...................................................... 136

Tutorial 28 – Ultrafast Demagnetisation ............................................................. 138

Tutorial 29 – Magneto-Optical Effect .................................................................. 142

Tutorial 30 – Spin-Wave Dispersion ................................................................... 143

Tutorial 31 – Two-Sublattice Model .................................................................... 146

Tutorial 32 – Exchange Bias ................................................................................ 152

Tutorial 33 – Magneto-Elastic Effect ................................................................... 153

Tutorial 34 – Atomistic Modelling ....................................................................... 156

4
User-Defined Text Equations .............................................................................. 157

Working with OVF2 Files ..................................................................................... 163

Materials Database ............................................................................................... 165

Differential Equations .......................................................................................... 168

Modules................................................................................................................. 180

Material Parameters ............................................................................................. 197

Commands – Essential ........................................................................................ 203

Commands – Important ....................................................................................... 204

Commands – Useful ............................................................................................. 205

Commands – All (Alphabetical) .......................................................................... 206

Selected Publications using Boris...................................................................... 255

5
Installation - Windows

An installer has been provided with the program and instructions therein should be
followed. The program is designed to run on Windows 7 and Windows 10, 64 bit
versions, and requires Microsoft Visual C++ 2017 Redistributable (x64) – included
with the installer. On Windows 10 the executable (Boris.exe) must be run in
compatibility mode – the installer sets this.

CUDA

To enable CUDA computations Boris requires a CUDA-enabled graphics card with

CUDA compute capability 5.0 or greater. The installer detects the CUDA compute
capability of the graphics card and launches the required program version. You
should always run the installed Boris.exe program, not the separate CUDA
versions found in the same directory.

The following architectures are supported by the installer package: Maxwell (sm_50,
sm_53), Pascal (sm_60, sm_61, and sm_62), Volta (sm_70, sm_72). The program
has not been tested on the Turing architecture (sm_75). If you have problems with
the program on this architecture please let me know ([email protected]).

Known Issues

Running DiagTrack and DPS (Diagnostic Policy Service) Windows services

can in certain cases result in very poor performance. In this case these
services should be stopped and disabled.

6
Installation – Linux-based OS

Extract the archive. On Linux-based OS the program needs to be compiled from

source using the provided makefile in the extracted BorisLin directory.

Make sure you have all the required updates and dependencies:

Updates:

1. Get latest g++ compiler: $ sudo apt install build-essential

2. Get OpenMP: $ sudo apt-get install libomp-dev
3. Get CUDA: $ sudo apt install nvidia-cuda-toolkit
4. Get SFML: $ sudo apt-get install libsfml-dev
5. Get FFTW3: Instructions at http://www.fftw.org/fftw2_doc/fftw_6.html

Open terminal and go to extracted BorisLin directory.

Step 1: Configuration.

$ make configure (arch=xx) (sprec=0/1)

Before compiling you need to set the correct CUDA architecture for your NVidia
GPU.

For a list of architectures and more details see: https://en.wikipedia.org/wiki/CUDA.

Possible values for arch are:

 arch=50 is required for Maxwell architecture; translates to

-arch=sm_50 in nvcc compilation.
 arch=60 is required for Pascal architecture; translates to
-arch=sm_60 in nvcc compilation.
 arch=70 is required for Volta (and Turing) architecture; translates to
-arch=sm_70 in nvcc compilation.

Example: $ make configure arch=70

If arch is not specified a default value of 50 is used.

7
You can also compile CUDA code with single or double precision floating point. The
default value, if not specified, is sprec=0 (single precision – recommended for most
users). If you have a GPU capable of handling double precision floating point
efficiently you can configure with sprec=1.

Step 2: Compilation.

$ make compile -j N

(replace N with the number of logical cores on your CPU for multi-processor
compilation, e.g. $ make compile -j 16)

Step 3: Installation.

$ make install

Run:

$ ./BorisLin

Notes:

In the current version Boris runs in text mode only on Linux-based OS, thus the
GRAPHICS 0 value needs to be kept in CompileFlags.h file. In a future version a
graphical interface will be ported to Linux also.

8
Overview

This manual contains a set of self-teaching tutorials that guide the user through most
of its functionality. The tutorials contain a number of exercises designed for users
without a background in micromagnetics, and may be skipped by more advanced
users. A number of examples that accompany the tutorials have also been provided
in the accompanying Examples folder.

You can use Tutorial 0 as a quick-start. This tutorial contains a number of Python
scripts as examples but doesn’t contain in-depth explanations.

Tutorials 1 to 7 cover the basics and it is recommended all users read through them.
After these you can skip to the required tutorials as needed. Tutorial 9 covers the
basics of automating simulations using Python and should be used as a starting
point if required. For the transport solver Tutorials 8 and 10 should be used as a
starting point. Tutorials 17 to 23 cover the spin transport solver. Tutorials 14 to 16
cover the heat solver.

The equations solved are given in the Differential Equations and Modules sections.
All material parameters used in these equations have been given in the Material
Parameters section.

A full list of commands has been provided in the Commands section in alphabetical
order. The most commonly used commands have also been outlined.

9
Tutorial 0 – Quick-Start

This tutorial contains a set of Python scripts for a selected number of

micromagnetics problems, and are intended as a crash-course for experienced users
who want to get things going very quickly. No detailed comments are provided,
except for comments in the Python scripts – if you want in-depth explanations you
need to read the other tutorials. All scripts are found in the Examples/Tutorial 0
folder.

The Python scripts below can be run either locally (connect as ‘localhost’ – this is the
default), or remotely (connect using ip address of machine running Boris).

For the Python scripts below you need the NetSocks.py module in the same
directory. Before executing the script Boris needs to be running; scripted
communication is enabled by default, but if you’ve disabled it you need to re-enable
it (use scriptserver 1 command).

The default program start-up state is a permalloy rectangle with dimensions 80 nm ×

80 nm × 10 nm, and cubic 5 nm cellsize. The demag, exchange, and Zeeman
modules are enabled. The LLG equation is set with the RKF45 evaluation method.
To restore the program to default state at any time use the default console
command (or alternatively send it as ns.default() from a Python script).

The default simulation stage is a Relax stage with |mxh| < 10-4 stopping condition
and no data saving condition.

Hysteresis Loop

import os
import sys
from NetSocks import NSClient

#setup communication with server.

ns = NSClient('localhost')

10
########################################

#the working directory : same as this script file

directory = os.path.dirname(sys.argv[0]) + "/"
#restore program to default state
ns.default()
ns.chdir(directory)

########################################

#This is based on Exercise 2.1, done entirely using a Python script

ns.meshrect([160e-9, 80e-9, 10e-9])

ns.cellsize([5e-9, 5e-9, 10e-9])

#setup two stages to sweep field up and down between -100 kA/m and +100kA/m in 100 steps, slightly off-axis.
#setstage sets a single stage, replacing the default stage
ns.setstage('Hxyz_seq')
ns.editstagevalue(0, [-100e3, 1e3, 0, +100e3, 1e3, 0, 100])
#add new stage
ns.addstage('Hxyz_seq')
ns.editstagevalue(1, [100e3, 1e3, 0, -100e3, 1e3, 0, 100])

#stop each field step using |mxh| < 10^-5 condition

ns.editstagestop(-1, 'mxh', 1e-5)
ns.editdatasave(-1, 'step')

#output data : applied field and average magnetisation

ns.setdata('Ha')
ns.adddata('<M>')
ns.savedatafile('hysteresis.txt')

#solve using LLGStatic equation (damping set to 1 and no precession term)

ns.setode('LLGStatic', 'RKF45')

#run program
ns.Run()

#output file has field (x, y, z components) in columns 0, 1, 2, and average magnetisation (x, y, z components) in
columns 3, 4, 5
hysteresis_data = ns.Get_Data_Columns('hysteresis.txt', [0, 3])
#plot Mx vs Hx
ns.Plot_Data(hysteresis_data[0], hysteresis_data[1], xlabel = 'H (A/m)', ylabel = 'M (A/m)', title = 'Hysteresis
Loop')

11
Domain Wall Movement

import os

import sys

from NetSocks import NSClient

import matplotlib.pyplot as plt

import numpy as np

#setup communication with server

ns = NSClient('localhost')

########################################

#the working directory : same as this script file

directory = os.path.dirname(sys.argv[0]) + "/"

#restore program to default state

ns.default()

ns.chdir(directory)

########################################

#This is based on Exercise 5.1, done entirely using a Python script

ns.meshrect([320e-9, 80e-9, 20e-9])

ns.cellsize([5e-9, 5e-9, 5e-9])

#setup the moving mesh algorithm for a transverse domain wall along the x axis:

ns.preparemovingmesh()

#relax dw in zero field to |mxh| < 10^-5

ns.editstagestop(0, 'mxh', 1e-5)

ns.Run()

#setup 2 field stages, each 5 ns long, but only second one saves data at 1 ps time intervals

ns.setstage('Hxyz')

ns.editstagestop(0, 'time', 5e-9)

ns.addstage('Hxyz')

ns.editstagestop(1, 'time', 5e-9)

ns.editdatasave(1, 'time', 1e-12)

#save time (s) and dw shift (m) data

ns.setdata('stime')

ns.adddata('dwshift')

ns.savedatafile('dwmovement_temp.txt')

#set fixed time-step RK4 method with 500fs time step

12
ns.setode('LLG', 'RK4')

ns.setdt(500e-15)

#save setup simulation file (next time you can just load it using ns.loadsim('dwmovement'))

ns.savesim('dwmovement')

########################################

Hrange = np.arange(100, 2400, 200)

dwvelocity = np.array([])

for H in Hrange:

ns.reset()

#first stage achieves steady state movement

ns.editstagevalue(0, H)

#second stage captures data

ns.editstagevalue(1, H)

ns.Run()

#process data to extract domain wall velocity

ns.dp_load('dwmovement_temp.txt', [0, 1, 0, 1])

ns.dp_replacerepeats(1)

dwdata = ns.dp_linreg(0, 1)

dwvelocity = np.append(dwvelocity, dwdata[0])

print('H (A/m) = %f, DW velocity (m/s) = %0.4f' % (H, dwdata[0]))

plt.axes(xlabel = 'H (A/m)', ylabel = 'DW Velocity (m/s)', title = 'DW Velocity and Walker Breakdown')

plt.plot(Hrange, dwvelocity, 'o-')

plt.show()

13
Anisotropic Magnetoresistance

import os

import sys

from NetSocks import NSClient

import matplotlib.pyplot as plt

import numpy as np

#setup communication with server

ns = NSClient('localhost')

########################################

#the working directory : same as this script file

directory = os.path.dirname(sys.argv[0]) + "/"

#restore program to default state

ns.default()

ns.chdir(directory)

########################################

#This is based on Exercise 8.3, done entirely using a Python script

ns.meshrect([160e-9, 80e-9, 10e-9])

ns.cellsize([5e-9, 5e-9, 5e-9])

#amr loop angle (deg.)

direction_deg = 5

ns.addmodule('permalloy', 'transport')

#set electrodes at x-axis ends with a 1 mV potential drop

ns.setdefaultelectrodes()

ns.setpotential(1e-3)

ns.setstage('Hpolar_seq')

ns.editstagevalue(0, [-100e3, 90, direction_deg, 100e3, 90, direction_deg, 200])

ns.editstagestop(0, 'mxh', 1e-6)

ns.editdatasave(0, 'step')

14
ns.addstage('Hpolar_seq')

ns.editstagevalue(1, [100e3, 90, direction_deg, -100e3, 90, direction_deg, 200])

ns.editstagestop(1, 'mxh', 1e-6)

ns.editdatasave(1, 'step')

ns.setangle(90, 180.0 + direction_deg)

#set amr percentage of 2%

ns.setparam('permalloy', 'amr', 2.0)

#save applied field (A/m) and resistance (Ohms)

ns.setdata('Ha')

ns.adddata('R')

ns.savedatafile('amr_rawdata.txt')

ns.setode('LLGStatic', 'SDesc')

ns.Run()

#load all columns from file (0, 1, 2, 3) into internal arrays (0, 1, 2, 4)

ns.dp_load('amr_rawdata.txt', [0, 1, 2, 3, 0, 1, 2, 4])

#get field strength along loop direction and save it in internal array 3

ns.dp_dotprod(0, np.cos(np.radians(direction_deg)), np.sin(np.radians(direction_deg)), 0, 3)

#save field strength and resistance in processed file, then plot it here

ns.dp_save('amr_loop.txt', [3, 4])

amr_data = ns.Get_Data_Columns('amr_loop.txt', [0, 1])

plt.axes(xlabel = 'H (A/m)', ylabel = 'R (Ohms)', title = 'AMR Loop')

plt.plot(amr_data[0], amr_data[1])

plt.show()

15
RKKY Simulation (multi-mesh demonstration)

import os

import sys

from NetSocks import NSClient

import matplotlib.pyplot as plt

import numpy as np

#setup communication with server

ns = NSClient('localhost')

########################################

#the working directory : same as this script file

directory = os.path.dirname(sys.argv[0]) + "/"

ns.default()

ns.chdir(directory)

########################################

direction_deg = 1.0

ns.meshrect([320e-9, 160e-9, 10e-9])

#shape mesh as an ellipse (mask file is stretched to mesh aspect ratios)

ns.loadmaskfile('Circle')

#add a new ferromagnetic mesh (permalloy by default) above the first one with 1 nm separation

#the 2 meshes still retain a cubic 5 nm cellsize

ns.addmesh('permalloy2', [0.0, 0.0, 11e-9, 320e-9, 160e-9, 31e-9])

ns.meshfocus('permalloy2')

ns.loadmaskfile('Circle')

#enable multilayered demag field calculation allowing exact and efficient calculation of demag fields, even though
the separation between meshes is 1 nm

ns.addmodule('supermesh', 'sdemag')

#enable RKKY coupling (surface exchange coupling) keeping default J1 (bilinear) and J2 (biquadratic) values

ns.addmodule('permalloy', 'surfexchange')

ns.addmodule('permalloy2', 'surfexchange')

#set field sequence to apply to both meshes (so set it to the supermesh)

ns.setstage('Hpolar_seq', 'supermesh')

ns.editstagevalue(0, [-300e3, 90, direction_deg, 300e3, 90, direction_deg, 300])

ns.editstagestop(0, 'mxh', 1e-5)

ns.editdatasave(0, 'step')

16
ns.addstage('Hpolar_seq', 'supermesh')

ns.editstagevalue(1, [300e3, 90, direction_deg, -300e3, 90, direction_deg, 300])

ns.editstagestop(1, 'mxh', 1e-5)

ns.editdatasave(1, 'step')

ns.setdata('Ha')

ns.adddata('<M>', 'permalloy')

ns.adddata('<M>', 'permalloy2')

ns.savedatafile('rkky_hysteresis.txt')

ns.setode('LLGStatic', 'SDesc')

ns.cuda(1)

ns.Run()

########################################

#process raw data into a hysteresis loop for the entire bilayer

u = [np.cos(np.radians(direction_deg)), np.sin(np.radians(direction_deg)), 0]

ns.dp_load('rkky_hysteresis', [0, 1, 2, 3, 4, 5, 6, 7, 8, 0, 1, 2, 3, 4, 5, 6, 7, 8])

ns.dp_dotprod(0, u[0], u[1], u[2], 10)

ns.dp_dotprod(3, u[0], u[1], u[2], 11)

ns.dp_dotprod(6, u[0], u[1], u[2], 12)

ns.dp_mul(11, 1.0/3)

ns.dp_mul(12, 2.0/3)

ns.dp_adddp(11, 12, 13)

ns.dp_save('rkky_hysteresis_loop.txt', [10, 13])

loop = ns.Get_Data_Columns('rkky_hysteresis_loop.txt', [0, 1])

plt.axes(xlabel = 'H (A/m)', ylabel = 'M (A/m)', title = 'RKKY Hysteresis Loop')

plt.plot(loop[0], loop[1])

plt.show()

17
Setting Shapes Programatically

import os

import sys

from NetSocks import NSClient

import numpy as np

from itertools import product

ns = NSClient('localhost')

########################################

directory = os.path.dirname(sys.argv[0]) + "/"

ns.default()

ns.chdir(directory)

########################################

#Mesh with Nxy cells along x and y, and Nz cells along z.

#Easier to generate mesh this way but the OVF2 file can be loaded in an arbitrarily shaped mesh in Boris
(mapped to dimensions).

Nxy, Nz = 32, 16

#M list to write in OVF2 file : has Nxy*Nxy*Nz cells, initialised with empty cells

M = [[0,0,0]] * Nxy**2*Nz

#setup hollow hemisphere values

origin = [Nxy/2, Nxy/2, Nz]

inner_ratio, outer_ratio = 0.75, 1.0

rad_inner, rad_outer = Nxy * inner_ratio / 2, Nxy * outer_ratio / 2

for i, j, k in product(range(Nxy), range(Nxy), range(Nz)):

rad = np.sqrt((i - origin[0])**2 + (j - origin[1])**2 + (k - origin[2])**2)

if rad >= rad_inner and rad <= rad_outer:

#Mark these cells as non-empty

M[i + j*Nxy + k*Nxy*Nxy] = [1,0,0]

#Write M to OVF2 file ready to load into Boris

fileName = 'HHemi.ovf'

#cellsize used to generate mesh rectangle

h = 5e-9

ns.Write_OVF2(fileName, M, [Nxy, Nxy, Nz], [0.0, 0.0, 0.0, Nxy*h, Nxy*h, Nz*h])

ns.meshrect([0.0, 0.0, 0.0, Nxy*h, Nxy*h, Nz*h])

ns.loadovf2mag(8e5, fileName)

18
Exchange Bias

import os

import sys

from NetSocks import NSClient

import matplotlib.pyplot as plt

#setup communication with server

ns = NSClient('localhost')

########################################

#the working directory : same as this script file

directory = os.path.dirname(sys.argv[0]) + "/"

#restore program to default state

ns.default()

ns.chdir(directory)

########################################

ns.setafmesh('Antiferromagnet', [320e-9, 320e-9, 10e-9])

ns.cellsize([5e-9, 5e-9, 5e-9])

ns.addmodule('Antiferromagnet', 'aniuni')

ns.addmodule('Antiferromagnet', 'surfexchange')

#set sub-lattice A magnetisation to result in biasing towards +ve side

ns.setangle(90, 180)

#Add Fe mesh on top of the antiferromagnet

ns.addmaterial('Fe', [0, 0, 10e-9, 320e-9, 320e-9, 12e-9])

ns.meshfocus('Fe')

#need smaller cellsize for Fe (in Boris meshes can be independently discretised)

ns.cellsize([2.5e-9, 2.5e-9, 2e-9])

ns.pbc('Fe', 'x', 10)

ns.pbc('Fe', 'y', 10)

ns.addmodule('Fe', 'anicubi')

#Now both the Antiferromagnet and Fe meshes have surfexchange module enabled, so exchange bias field will
be included in computations

19
ns.addmodule('Fe', 'surfexchange')

#set bilinear surface exchange coupling value - exchange bias is proportional to this

ns.setparam('Fe', 'J1', 0.2e-3)

ns.setode('LLGStatic', 'RKF45')

ns.setstage('Hpolar_seq', 'supermesh')

ns.editstagevalue(0, [-50e3, 90, 5, 100e3, 90, 5, 50])

ns.editstagestop(0, 'mxh', 1e-4)

ns.editdatasave(0, 'step')

ns.addstage('Hpolar_seq', 'supermesh')

ns.editstagevalue(1, [100e3, 90, 5, -50e3, 90, 5, 50])

ns.editstagestop(1, 'mxh', 1e-4)

ns.editdatasave(1, 'step')

ns.setdata('Ha')

ns.adddata('<M>', 'Fe')

ns.savedatafile('exchangebias.txt')

ns.cuda(1)

ns.Run()

#we should really project along the 5 degree direction, but will keep this simple

data = ns.Get_Data_Columns('exchangebias.txt', [0, 3])

plt.axes(xlabel = 'H (A/m)', ylabel = 'M (A/m)', title = 'Exchange bias')

plt.plot(data[0], data[1])

plt.show()

20
Ultrafast Demagnetisation and Skyrmion Creation in a Co/Pt/SiO2 Trilayer
(Advanced)

import os

import sys

from NetSocks import NSClient

import numpy as np

import matplotlib.pyplot as plt

ns = NSClient('localhost')

########################################

directory = os.path.dirname(sys.argv[0]) + "/"

ns.default()

ns.chdir(directory)

########################################

ns.setode('sLLB', 'TEuler')

#Co layer

ns.setmaterial('Co/Pt', [512e-9, 512e-9, 2e-9])

ns.cellsize([1e-9, 1e-9, 2e-9])

ns.scellsize([4e-9, 4e-9, 2e-9])

ns.setdtstoch(20e-15)

ns.addmodule('Co/Pt', 'iDMexchange')

ns.addmodule('Co/Pt', 'aniuni')

ns.addmodule('Co/Pt', 'heat')

ns.tcellsize([2e-9, 2e-9, 2e-9])

ns.tmodel(2, 'Co/Pt')

ns.curietemperature(500)

ns.pbc('Co/Pt', 'x', 10)

ns.pbc('Co/Pt', 'y', 10)

#Pt layer

ns.addmaterial('Pt', [0.0, 0.0, -8e-9, 512e-9, 512e-9, 0.0])

ns.meshfocus('Pt')

ns.addmodule('Pt', 'heat')

ns.tcellsize([4e-9, 4e-9, 4e-9])

ns.tmodel(2, 'Pt')

#SiO2 layer

ns.addmaterial('SiO2', [0.0, 0.0, -48e-9, 512e-9, 512e-9, -8e-9])

21
ns.meshfocus('SiO2')

ns.addmodule('SiO2', 'heat')

ns.tcellsize([8e-9, 8e-9, 8e-9])

#general settings

ns.setangle(0, 0)

ns.setfield(100e3, 0, 0)

ns.temperature(300)

#simulation stage

ns.setstage('Qequation', 'Co/Pt')

ns.editstagevalue(0, 'Q0 * exp(-sqrt((x/Lx - 0.5)^2 + (y/Ly - 0.5)^2) / ((d0/Lx)^2/(4*ln(2)))) * exp(-(t-

2*tau)^2/(tau^2/(4*ln(2))))')

ns.equationconstants('d0', 400e-9)

ns.equationconstants('tau', 100e-15)

ns.equationconstants('Q0', 4e21)

#output data

ns.setdata('time')

ns.adddata('Q_topo', 'Co/Pt')

ns.adddata('<T>', 'Co/Pt', [255e-9, 255e-9, 0.0, 256e-9, 256e-9, 2e-9])

ns.editdatasave(0, 'time', 10e-15)

ns.savedatafile('ufsky.txt')

#set-up display

ns.meshfocus('Co/Pt')

ns.display('M', 'Co/Pt')

ns.vecrep('Co/Pt', 3)

ns.cuda(1)

#next time you can just load this and run it

ns.savesim('ufsky_fm')

#0 to 10ps

ns.editstagestop(0, 'time', 10e-12)

ns.setdt(1e-15)

ns.setheatdt(1e-15)

ns.Run()

#10ps to 20ps

ns.editstagestop(0, 'time', 20e-12)

ns.setdt(2e-15)

22
ns.setheatdt(2e-15)

ns.Run()

#20ps to 60ps

ns.editstagestop(0, 'time', 60e-12)

#mid-simulation re-meshing! 1nm cellsize only needed around the Curie temperature.

#finely tuned simulations of this type means you can run thousands of these events in a reasonable time-scale
and still keep accuracy

ns.cellsize([2e-9, 2e-9, 2e-9])

ns.setdt(10e-15)

ns.setheatdt(2e-15)

ns.Run()

#60ps to 800ps

ns.editstagestop(0, 'time', 800e-12)

ns.tcellsize([4e-9, 4e-9, 1e-9])

ns.setdt(50e-15)

#could do with Crank-Nicolson method in next version

ns.setheatdt(5e-15)

ns.editdatasave(0, 'time', 250e-15)

ns.Run()

#now plot |Q| as a function of time

data = ns.Get_Data_Columns('ufsky.txt', [0, 1])

time_ps = [t/1e-12 for t in data[0]]

Qmod = [np.abs(Qval) for Qval in data[1]]

plt.axes(xlabel = 'Time (ps)', ylabel = '|Q|')

plt.xscale('log')

plt.yscale('log')

plt.plot(time_ps, Qmod)

plt.xlim(0.1)

plt.ylim(1e-3)

plt.savefig('ufsky_plot.png', dpi = 600)

plt.show()

23
Tutorial 1 – Introduction

Basics

All commands are entered using the console (top-left black box) in Figure 1.1. Each
console command has a case-sensitive syntax and may have a number of
parameters separated by spaces. If a command is entered with wrong parameters a
help prompt will be displayed explaining the command and full syntax. Alternatively a
command may be immediately preceded by a question mark in order to display the
command help. For example try it for the run command:

?run

Note, the program auto-completes commands entered out of the list of possible
inputs – and equally stops any wrong inputs from being entered.

The main display shows the magnetization configuration (other vector and scalar
quantities may be displayed, but magnetization is the default setting).

Figure 1.1 – Boris interface

24
The magnetization display may be controlled using the mouse: left-click and drag to
re-position the display mesh, middle-click and move mouse to rotate the camera
view about the center of the displayed (focused) mesh, right-click and move mouse
up/down to zoom in/out, or left/right to rotate the camera about its axis; finally the
wheel may be used to set the coarseness of magnetization representation: for large
mesh dimensions, each arrow represents an average of the magnetization in that
area.
Various simulation data may be displayed in the data box (top-right box) for
convenience – more on these later. The displayed windows may be resized by
dragging their outlines – the outline appears if you hover the mouse over the edge of
a window.

The magnetization display can be reset to the default view using:

center

Simulation Mesh Control

To display the current problem size enter the following command (see Figure 1.2 for
expected output):

mesh

Figure 1.2 – Default mesh configuration

The simulation space consists of one or more named meshes – the default
configuration consists of a single ferromagnetic mesh named permalloy. This has a
rectangle with lower-left corner coordinates of (0, 0, 0) and upper-right corner
coordinates of (80 nm, 80 nm, 10 nm). The magnetic discretization cellsize is a
rectangular prism with dimensions (dx, dy, dz) = (5 nm, 5 nm, 5 nm) – a cubic cellsize
by default. Thus the permalloy mesh is discretized with the integer number of cells
(16, 16, 2).

25
To adjust the mesh dimensions you can use the meshrect command. An easy way
to bring up this command, with current fields already entered, is to double-click on
the outlined text containing the mesh rectangle dimensions:

This type of outlined text is a special console text called an interactive console
object, allowing a number of user interactions depending on the particular object,
including left or right click, double-click, or drag. The text is also automatically
updated to display currently set values. You can find out what an interactive object
does by using shift-click.

Try to resize the permalloy mesh so it has the dimensions (300 nm, 100 nm, 15 nm).
Values may be entered without specifying the units, in which case the applicable S.I.
unit is assumed, or the applicable unit may be entered together with its magnitude
specifier (e.g. for a meter the currently available units are designated as am, fm, pm,
nm, um, mm, m, km, Mm, Gm, Tm, Pm). If entering the unit, do not leave a space
between the number and unit.

The magnetic cellsize may be adjusted by double-clicking on the magnetic cell

interactive object, which brings up the cellsize command. Try to adjust the cellsize
so it has dimensions (6 nm, 6 nm, 5 nm). After changing the cellsize its dimensions
are automatically adjusted in order to satisfy the requirement of integer number of
discretization cells in each dimension.

Similarly the permalloy mesh may be renamed by double-clicking on the mesh name
interactive object, which brings up the renamemesh command.

In Figure 1.2 you can also see an entry for the supermesh. Its rectangle is not
controlled directly, but depends on the currently set meshes (however you can adjust
the supermesh cellsize). The magnetic supermesh is the smallest simulation space
containing all the currently set magnetic meshes and is useful to compute long-range
interactions over several independently discretized meshes (e.g. supermesh
demagnetizing field) – more on this in a dedicated tutorial.

26
Basic Simulation Control

The simulation is started and stopped using:

run
stop

The stop command simply pauses the simulation without resetting it. To continue
from the stop point simply type run again. To reset the simulation use:

reset

The display refresh frequency can be set using (iter is the number of iterations –
remember you can query to command for full details: ?iterupdate):

iterupdate iter

The simulation may be saved at any point using (do not use a termination, the .bsm
termination is added by default):

savesim filename

If a directory path is not specified, the default directory path is used. To set a default
directory use:

chdir directory

To load a previously saved simulation use:

loadsim filename

Alternatively a simulation file may be dragged into the console area. At any point you
can return to the default program state by using:

27
default

A uniform magnetization configuration can be set using the following (theta is the
polar angle, phi is the azimuthal angle in spherical polar coordinates):

setangle theta phi

A uniform magnetic field can be set using (again use spherical polar coordinates):

setfield Hmag Htheta Hphi

Simulation Modules

Simulation modules typically correspond to effective field terms. These can be

managed using interactive objects by typing the following command:

modules

The default configuration includes the demagnetizing field (demag), direct exchange
interaction (exchange), and applied field (zeeman) – see Figure 1.3.

Figure 1.3 – Default simulation modules

Currently added modules are displayed in green. To add or remove a module left or
right-click on the respective interactive object. Individual modules will be explored in
future tutorials.

Material Parameters

Default simulation parameters are set for permalloy (Ni80Fe20), as Ms = 8e5 A/m,
A = 1.3e-11 J/m, and α = 0.02. To see a list of currently set parameter values use the
command:

28
params

Values may be modified by double-clicking on the respective interactive objects. By

default parameters are constant for each mesh, however they can be assigned
temperature and spatial dependence for advanced simulations (more on this later).

Simulation Flow

A basic simulation flow can be programmed by setting a number of stages. Each

stage has an identifier, parameters depending on the identifier, a stopping condition
and data save condition. To show the currently set simulation stages use:

stages

By default the Relax stage type is used (no simulation values changed), with a
stopping condition based on the normalized torque |mxh| < 10-4 (mxh: 0.0001), and
no data saving configured. New stages may be added by double-clicking on the
interactive objects at the bottom – see Figure 1.4.

Figure 1.4 – Simulation stage types

You can delete added stages by right-clicking on them, change the stage type by
double-clicking and editing, and re-arrange the stage order by dragging.

Available stopping conditions are: nostop, iter (stop after a number of iterations),
mxh (stop when |mxh| falls below the set threshold), dmdt (stop when the normalized
|∂m/∂t| value falls below the set threshold), or time (stop after an elapsed simulation
time). When a stage reaches the stopping condition the next stage starts or the
simulation finishes.

Some stage types are broken down into several sub-stages (referred to as steps), for
example a field sequence using Hxyz_seq. Try to add a field sequence stage by

29
double-clicking on the Hxyz_seq interactive object. The default parameters for a field
sequence are shown in Figure 1.5. This consists of a field sequence starting from a
field of -100 kA/m along the x-axis, and stopping at +100 kA/m along the x-axis. The
sequence consists of 100 field steps, thus the field step is 2 kA/m. The simulation
proceeds to the next step when the |mxh| < 10-4 stopping condition is satisfied.

Figure 1.5 – Default parameters for the Hxyz_seq stage type

Exercise 1.1

Set a 160×80×5 nm permalloy mesh (with cubic cellsize of 5 nm) starting from a
saturated magnetization state along the negative x-axis direction. Set a field
sequence from -60 kA/m to +60 kA/m along the x-axis using a step of 2 kA/m and
mxh stopping condition of 10-4 (typically this threshold is too high, it should be 10-5 or
even lower for an accurate simulation depending on the problem, but this will speed
up the exercise).

Display the applied field and average magnetization in the data box. To do this use
the command:

data

You will see a list of interactive console objects representing possible output data
which can be displayed in the data box or saved to a file. For now just display the
applied field (Ha) and average magnetization (<M>) by right-clicking on the
interactive objects (or dragging them to the data box).

Run the simulation - the magnetization should switch during this field sequence.

30
Tutorial 2 – Data Output

Saving Numerical Data

In order to save numerical simulation data (automated saving of images will be

explored in a future tutorial) you need to set a data saving file, a list of output data,
and a saving schedule. To set output data and a save file use the data command.

Figure 2.1 – Default output data

The default output data file is called out_data.txt and its name may be modified by
double-clicking on the interactive object. The default saving directory is the path to
the program executable file and may be modified by double-clicking on the
interactive object.

The default output data includes sstep (stage and step), iter (iteration), time
(simulation time), Ha (applied field), and <M> (average magnetization). This is the
order the output data will appear in the output file as numerical columns. The order
may be modified by dragging the respective interactive objects in the list of output
data. New output data may be added by double-clicking on the interactive objects at
the bottom, and set output data may be deleted by right-clicking on the respective
interactive objects in the output list.

Some output data (such as ha and <M>) may be saved in a particular mesh – in this
case the permalloy mesh which is specified using the notation <permalloy>, whilst
other output data do not depend on any particular mesh. Some output data (such as

31
<M>) may also be saved in a particular rectangle of the named mesh (the rectangle
is relative to the named mesh) – by default the entire mesh rectangle is saved, but
this can be modified by double-clicking on the respective interactive object and
editing.

Finally, a saving schedule may be set in the simulation stages: use the stages
command. Each stage has a list of possible saving conditions: none (default – do not
save), stage (save at the end of the stage), step (save at the end of each step in the
current stage), iter (save every n iterations), and time (save every t simulation
seconds); for iter and time the parameters may be edited by double-clicking the
respective interactive objects.

Exercise 2.1

Set a 160×80×10 nm permalloy mesh (with cubic cellsize of 5 nm) starting from a
saturated magnetization state along the negative x-axis direction. Set a field
sequence from -100 kA/m to +100 kA/m and then back to -100 kA/m along the x-axis
using a step of 2 kA/m and mxh stopping condition of 10-4.

Configure the simulation so it saves output data for a hysteresis loop (applied field
and average magnetization saved after every step).

Before running the simulation save the simulation file. Once the simulation file is
saved using a specified name (and directory path if needed), the next time you don’t
need to specify the file name – simply use savesim without a file name and the
previously used file name will be saved. Note, correct commands previously entered
in the console can be recalled using the arrow keys (invalid commands are not
saved). You can also use Ctrl^v to paste text in the console.

Run the simulation and plot the hysteresis loop (magnetization along the applied field
direction) at the end.

32
Figure 2.2 – Hysteresis loop obtained in exercise 2.1

Further Data Box Control

As introduced in the previous tutorial, output data may also be displayed in the data
box for convenience. The possible output data may be listed as interactive objects by
using the data command, and the listed interactive objects may be displayed in the
data box by dragging them there, or right-clicking on them. Data box entries may be
removed by right-clicking on them in the data box, and they may be re-arranged by
dragging them.

If you just want to quickly see the current values of particular data without displaying
them in the data box, bring up an interactive object list using the showdata
command and double-click on the respective interactive objects.

33
Exercise 2.2

In this exercise you will run the MAG standard problem #4:
https://www.ctcms.nist.gov/~rdm/std4/spec4.html

a) For this problem we need a permalloy mesh with dimensions 500x125x3 nm.
First initialize the magnetization configuration to a so-called s-state: this may
be obtained by reducing a large applied field to zero along the [1,1,1]
direction. For example set a field sequence starting from 1 MA/m along the
[1,1,1] direction, reducing to zero in 20 steps – you should use the stricter
|mxh| < 10-5 condition this time. Save an image of the obtained magnetization
configuration – see Figure 2.3.

The easiest way to setup Exercise 2.2a is to use a polar field sequence: Hpolar_seq.
This specifies the starting and ending field values using polar coordinates:
magnitude, polar angle and azimuthal angle – thus a starting field of 1 MA/m along
the [1,1,1] direction would be specified (roughly) as 1MA/m, 55, 45. Make sure to
specify the ending field value as 0, 55, 45 to keep the field values in the sequence
along the same direction. You should also set the starting magnetization state along
the [1,1,1] direction: setangle 55 45.

To save an image of the currently displayed mesh, use the command:

savemeshimage (directory\)filename

Figure 2.3 – Starting s-state for micromagnetics standard problem #4

34
Exercise 2.2 continued

b) Starting from the s-state, apply a fixed field with magnetic flux density (B-field)
of 25 mT directed 170 counterclockwise from the x-axis in the x-y plane.
Simulate the switching event for a duration of 5 ns, saving output data (in
particular the average magnetization and simulation time are required) every
5 ps. Plot the 3 components of magnetization against time. Remember to
save the simulation before starting it. How do these results compare with
published solutions ?

(see https://www.ctcms.nist.gov/~rdm/std4/results.html)

c) Repeat part b) but this time for a B-field of 36 mT directed 190 degress
counterclockwise from the x-axis in the x-y plane.

d) For parts b) and c) obtain images of the magnetization configuration when the
average magnetization (x component) first crosses zero – use the output data
to determine the time when this occurs, then run the simulation to stop at this
particular time.

35
Figure 2.4 – a) Results obtained after running the MAG standard problem #4 with
field 1, and b) magnetization configuration when the average magnetization (x
component) first crosses zero.

a)

b)

36
Figure 2.5 – a) Results obtained after running the MAG standard problem #4 with
field 2, and b) magnetization configuration when the average magnetization (x
component) first crosses zero.

a)

b)

Making a video from an image sequence

A video may be encoded from a sequence of .png files (e.g. as produced from a
simulation with an image saving schedule) – this functionality is built into the
program for convenience; for advanced image processing you should use an
external program. To produce a video file from an image sequence, use:

makevideo (directory\)filebase fps quality

37
This makes a video from all .png files which start with the filebase name, including
the directory, at the given fps (frames per second). The quality parameter sets the
bit-rate of the output video: 0 for worst quality but smallest size, 5 for best quality but
largest size. For the makevideo command, the files are sorted by their creation time,
not alphabetically.

To enable mesh image saving, use the data command then click on the respective
interactive object. You can also edit the mesh image filebase name. The mesh
images are saved using the same save conditions as output data.

Exercise 2.3

For the switching event in exercise 2.2b, set the problem to save mesh images every
10 ps; also reduce the simulation time to 3 ns and disable data saving. Make a video
of the switching event (40 fps and quality level 3 works well).

If you want to capture only a part of the mesh display window you can set cropping
factors. These are specified as normalized values and are applied whenever an
image is saved (either manually or during a simulation). This is set using:

imagecropping left bottom right top

The left-bottom of the mesh display is (0, 0), whilst the right-top of the mesh display
is (1,1).

38
Tutorial 3 – Further Data Output

Exercise 3.1

In this exercise you will compare coercive fields obtained using the full
micromagnetics model with the predictions of the simpler Stoner-Wohlfarth model.

a) Simulate an in-plane hysteresis loop along a 10 direction in a permalloy

rectangle with dimensions 250x50x5 nm using the full micromagnetics model
(demag, exch, and zeeman modules for permalloy) and obtain the coercive
field. In order to speed up the simulation you only need to simulate one
branch of the hysteresis loop, e.g. negative to positive field only, and you
should also set a coarse field step up to zero field (e.g. 10 kA/m), then a fine
field step in order to obtain a more accurate switching field value (use a 500
A/m fine field step or less); use the mxh stopping condition with a 10-5
threshold.

Solution: use two polar field sequences (Hpolar_seq) along the (polar, azimuthal) =
(90, 10) direction. Start at -200 kA/m and finish at 60 kA/m. This range is just
enough to start from a saturated state and capture the switching field.

i.e. : 1) Hpolar_seq -200kA/m 90 10 0kA/m 90 10 20, 2) Hpolar_seq 0kA/m 90 10

60kA/m 90 10 120

b) Compute the anisotropy energy density (shape anisotropy) and hence obtain
the switching field predicted by the Stoner-Wohlfarth model.

Note, the Stoner-Wohlfarth model predicts the switching field:

1 t2  t4
, where t  3 tan  
2Ku
HS 
0 M S 1 t 2

39
and  is the angle between the applied field and anisotropy easy axis (formula
applicable for  between 0 and 45).

To compute the energy density for a given magnetization orientation, set the
magnetization orientation (setangle) and calculate the energy density terms using
the command:

computefields

This command runs the simulation for a single iteration and does not advance the
simulation time – only the currently set simulation modules are refreshed. After
running this command the required energy density term (e_demag) will be available.

c) For the same geometry and applied field direction obtain the hysteresis loop
using the Stoner-Wohlfarth model. Does the coercive field agree with that
obtained in part b) ?

To run this you will need to use the demag_N module instead of the full
demag module; the exchange module (exch) is not needed.

The demag_N module computes the demagnetizing field using the simple
approximation Hd, i = -Ni Mi (i = x, y, z). You will need to enter correct values
for the demagnetizing factors Nx and Ny (remembering that Nx + Ny + Nz = 1).
These can be entered using the command params, then editing the values
under the Nxy interactive object.

Calculate Nx, Ny and Nz directly from the demagnetizing field (obtained using
the full micromagnetics model) and also using the demagnetizing energy
density values obtained in part b). Do the values agree, and does the
relationship Nx + Ny + Nz = 1 hold?

40
To obtain the demagnetizing field you will need to update the field using the
computefields command with only the demag module enabled. After this you can
display the demagnetizing field using the command:

display

Using this command brings up a list of interactive objects with display options. Click
on the Heff option under the permalloy mesh. This will display the computed effective
field. Using the average effective field value you can obtain a value for the
demagnetizing factor along the set magnetization direction using the expression Hd =
-N M.

In order to obtain the average value of the demagnetizing field you can use the
command:

dp_averagemeshrect (sx sy sz ex ey ez)

This command returns the average value for the displayed quantity in the currently
focused mesh (the permalloy mesh in this case) when used without parameters. The
parameters specify a mesh rectangle (start and end Cartesian coordinates) which is
relative to the currently focused mesh.

d) Repeat this exercise using the 100x25x5 nm.

e) Repeat this exercise using the 50x25x3 nm.

Another way to calculate the demagnetizing factors is to use the formula:

0
d   M.H d
2

Thus for M along i = x, y, z, you can obtain e_demag, then use:

0
 d ,i  N i M S2 (i  x, y, z )
2

41
Exercise 3.2

Here you will obtain the magnetization dynamics during a switching event and
investigate the effect of the cellsize value on the simulation.

a) Set a 320x160x10 nm permalloy rectangle with magnetization along the

length of the rectangle (set the magnetization towards the left, thus blue
coloured). Obtain the stable magnetization configuration at zero field by
reducing the magnetic field from a large saturation value along x to zero in a
number of steps. (e.g. from -50 kA/m to 0). For now use a cellsize of 5 nm.

b) Starting from the magnetization configuration set-up in part a) set a single

stage where you apply a large field along the x direction, opposing the
magnetization – use 50 kA/m. As stopping condition using a time interval of 4
ns. Set a saving schedule to save the simulation time and average
magnetization components every 10 ps. For now use a cellsize of 5 nm. From
the saved data plot <Mx> and <My> versus time.

c) Repeat the simulation in b) with cellsize values of 10 nm and 2.5 nm. Plot
<Mx> and <My> versus stage time for the 3 cellsize values. How do the results
compare ? Which cellsize would you recommend to use ?

42
Tutorial 4 – Domain Walls

Generating Domain Walls

An idealized domain wall (using a tanh profile) along the x direction can be
generated using:

dwall longitudinal transverse width position

For an in-plane domain wall the longitudinal parameter determines if the wall is
head-to-head (longitudinal = x) or tail-to-tail (longitudinal = -x); Bloch walls may be
generated using z or –z for the longitudinal component. The transverse parameter
determines the rotation direction through the wall – see Figure 4.1 for examples. The
width value is the total domain wall width and position is the starting left-hand-side
coordinate of the wall (along the x axis), relative to the focused mesh rectangle.

Figure 4.1 – Domain walls generated using the dwall command for a mesh with
240nm length, and:

a) in-plane head-to-head transverse domain wall as dwall x y 240nm 0

43
b) in-plane tail-to-tail transverse domain wall as dwall -x -y 240nm 0

c) Bloch domain wall as dwall z y 240nm 0

d) Néel domain wall as dwall z x 240nm 0

44
Exercise 4.1

Set a mesh for a permalloy rectangle of dimensions 640x80x5 nm with 5 nm cellsize.

Generate a head-to-head domain wall over this wire. Run the simulation without a
stopping condition and observe how the domain wall is relaxed. You will observe the
domain wall does not remain in the center, but eventually drifts towards one side
until it is expelled at one of the edges. Can you explain why this happens?

Setting Stray Fields

For domain wall mobility calculations and domain wall configuration relaxation
problems in very long wires, it is possible to extend the wires outside of the
ferromagnetic mesh by using external uniformly magnetized magnetic bodies, and to
calculate the stray field inside the mesh, thereby allowing a smaller mesh size – this
eliminates the domain wall drift problem noted in Exercise 4.1. This is done by
adding dipole meshes at the left and right-hand-side of the permalloy mesh with
magnetization direction set as a continuation of the magnetization inside the
permalloy mesh, and enabling the strayfield module.

To add a dipole mesh use the following:

adddipole name rectangle

A dipole mesh has a uniform magnetization orientation which is not evolved by the
ODE solver but may be handled in a similar manner to ferromagnetic meshes (such
as the permalloy mesh). Thus modules may be added for computation (modules),
mesh parameters edited (params), quantities displayed (display), etc.

You should also exchange couple the ends of the wires to the dipole meshes, thus
completing the approximation of a long wire with a domain wall in the center. To
enable this use the command:

coupletodipoles

45
Click on the interactive object to enable exchange coupling to the On state – with this
flag turned on all magnetic cells in a ferromagnetic meshes, at the interface with a
dipole mesh, are exchange coupled to the fixed dipole magnetization direction.

Exercise 4.2

a) Set two dipole meshes to the left and right of the permalloy mesh from the
previous exercise, with lengths of 2.56 µm (but same width and thickness).
These dipole meshes should now be visible when using the mesh command
– see Figure 4.2. Set their magnetization orientation in order to extend the
head-to-head domain wall configuration from Exercise 4.1 (use setangle
remembering to specify the mesh name – see ?setangle for details) – note,
the dipole meshes do not display anything by default, you will need to use the
display command and click the M interactive object in order to see their
magnetization orientation.

Run the simulation to relax the domain wall configuration to |mxh| < 10-5. What
does the stray field from the dipole meshes look like? (use the display
command)

Figure 4.2 – Configuration of dipole meshes for exercise 4.2

b) Change the permalloy mesh from part a) to a new size of 640x160x30 nm,
making sure the dipole meshes are also scaled accordingly (to 160nm width
and 30 nm thickness). For the purposes of this exercise you may use a 2D
approximation by setting a cellsize of 5x5x30 nm.

Run the simulation to relax the domain wall configuration to |mxh| < 10-5. What
type of domain wall results?

46
When changing mesh dimensions with multiple meshes added to the simulation,
there are two options available: i) change just the mesh rectangle required, ii)
change the mesh rectangle required and resize/translate all other meshes in
proportion. To change the behaviour of the program use the following command and
click the interactive object to the On state:

scalemeshrects

With multiple meshes, clicking on the mesh name interactive object (e.g. as
displayed using the mesh command) changes the display focus to that mesh and
resets camera orientation – try it. To quickly focus on a mesh without changing the
camera orientation you can double-click on a mesh in the display window.

Exercise 4.3

In this exercise you will calculate the domain wall width for a symmetric transverse
domain wall as a function of wire width and compare it to the values obtained using
the domain wall formula (A is the exchange stiffness – see the params command –
and Ku is the anisotropy energy density).

A
 dw  
Ku

a) Relax domain walls as a function of wire width for a permalloy mesh with
dimensions 640 x Width x 5nm, where Width ranges from 40 nm up to 160 nm
(simulate at least 4 different values of width). Obtain the domain wall width
defined as the half-Ms width value – i.e. the distance it takes for the
longitudinal component to change from +MS/2 to –MS/2 for a head-to-head
domain wall – and compare it to the value obtained using the formula above
(when calculating Ku remember the longitudinal demagnetizing energy is
assumed to be negligible as for a very long wire)

47
To obtain the domain wall profile you can use the following command:

dp_getprofile start end dp_index

The above command saves numerical data from the currently displayed mesh
quantities (magnetization in this case) along a line starting from the start up to the
end Cartesian coordinates (absolute position values, i.e. not relative to any mesh).
The data is saved in internal data processing arrays – more on these in a separate
tutorial. For now just obtain a magnetization profile through the middle of the wire as
(e.g. for the 80nm wire width): dp_getprofile 0 40nm 0 640nm 40nm 0 0.

The magnetization components are saved in the data processing arrays with indexes
starting at 1 (so 1, 2, and 3), whilst data processing array 0 contains the position
value. These can be saved to a file as numerical columns using the dp_save
command as dp_save (directory\)filename.txt 0 1 2 3. The file will contain the 4
columns as position along the profile (so x coordinate), Mx, My, Mz.

b) Repeat the exercise for a thickness of 10nm using both a 3D simulation (cubic
cellsize of 5x5x5 nm) and a 2D approximation (cellsize of 5x5x10 nm). How
do the width values compare to those predicted by the formula and are the
results obtained using the 2D and 3D model similar?

48
Tutorial 5 – Domain Wall Movement and Data Processing

In this tutorial you will learn how to obtain a domain wall field-driven mobility curve.
In order to simulate domain wall movement, in addition to setting up a domain wall
and dipole meshes as in the previous exercise, the moving mesh algorithm must be
enabled by setting a “triggering mesh” as:

movingmesh mesh_name

When moving mesh is enabled the magnetization is shifted either to the left or to the
right by one notch at a time in order to keep the average x component of
magnetization in the triggering mesh, <Mx>, within set boundaries. This keeps the
domain wall roughly in the centre of the mesh. When the mesh is shifted to the left or
to the right, the data parameter dwshift is changed. This data parameter is available
for saving to file – see list of data parameters using the data command, as discussed
previously. Note, this can also be displayed in the console using showdata dwshift,
or displayed in the data box. By saving the simulation time and domain wall shift, the
domain wall velocity can be calculated using linear regression.

Since this type of computation is common, there is a shortcut command which sets-
up everything required (adding dipole meshes with exchange coupling to the
ferromagnetic mesh, enabling stray field computation, setting a domain wall and a
triggering mesh):

preparemovingmesh (meshname)

Exercise 5.1

a) Set a 320 × 80 × 20 nm permalloy rectangle with 5x5x20 nm cellsize (2D

problem). Enable the moving mesh algorithm and let the domain wall relax in
zero field.

b) Set a simulation stage with a field sequence starting from 500 A/m to 2000
A/m in 500 A/m steps, keeping each field step for exactly 2 ns. Set a data

49
 save file, and make sure you save the stage time and dwshift parameters
every 50 ps.

c) For each field step extract the gradient of the dwshift vs time and plot the wall
velocity as a function of field. How does the velocity compare with that
predicted by the formula below? (Ku is the in-plane anisotropy energy density)

 A
vdw  H , where    0  e  221276 (m/As)
 Ku

Console Data Processing

There are a number of built-in commands which allow for a number of operations to
be performed on data processing arrays.

First of all, it is possible to load tab-spaced data from a file (such as the data files
produced by a Boris simulation) into the internal data processing arrays. This is done
using the following command:

dp_load filename filecol1 … dp_arr1 …

The above command loads entire columns from the specified file. Thus if the file has
a number of tab-spaced data columns, we can load the column with number filecol1
from the file into the internal data processing array with number dp_arr1 (these
indexes are numbered from 0 up). Multiple columns can be loaded in one command.

Some common data processing commands are listed below.

To multiply a data processing array by a constant value use the following command:

dp_mul dp_source value dp_dest

50
The above command multiplies the data processing array with index dp_source by
the specified value and stores the result in the dp_dest data processing array (this
can be the same as dp_source). This can be used to normalize data (e.g. a
hysteresis loop).

We can use the built-in linear regression command to extract the velocity values:

dp_linreg dp_x dp_y (dp_z dp_out)

The above command performs linear regression on the data stored in dp_x and
dp_y data processing arrays and outputs the extracted gradient values and
intercepts together with their uncertainties. If dp_z is specified multiple linear
regressions are performed by using the values in the dp_z array to identify adjacent
points to be included in a single linear regression; e.g. dp_z would contain the
applied field values. In this case the outputs are placed in 5 data processing arrays
starting at dp_out as follows: 1) unique dp_z values, 2) gradient, 3) gradient error, 4)
intercept, 5) intercept error.

We can also save our processed data, e.g. the domain wall velocity curve, using:

dp_save filename dp_arr1 …

Exercise 5.2

Use the console data processing commands to process the output data from
Exercise 5.1 and save a domain wall velocity curve.

Other notable commands include:

dp_coercivity dp_x dp_y

dp_remanence dp_x dp_y

These commands can be used on data from a simulated hysteresis loop in order to
extract coercivity and remanence values.
51
The data processing arrays may be cleared using:

dp_clear dp_arr1 …

This clears data in the specified data processing arrays. If no parameters are
included all data processing arrays are cleared.

Exercise 5.3

Continuing Exercise 1, find the Walker breakdown threshold with a resolution of 100
A/m starting at 100 A/m. Compare the velocity values with that predicted by the
formula in Exercise 5.1c for the steady domain wall movement regime. What is the
Walker breakdown threshold?

Exercise 5.4

Calculate the field-driven domain wall mobility curve for permalloy, with a resolution
of 100 A/m, as a function of Gilbert damping, for values of damping 0.005, 0.01 and
0.015. How does the Walker breakdown threshold compare with the value predicted
by the formula below? (Ku,op is the out-of-plane anisotropy energy density)

 K u ,op
HW  ( A / m)
2 0 M S

52
Tutorial 6 – ODE Control and Setting Shapes

Setting an ODE solver

The differential equation to solve and its evaluation method is configured using the
following command:

ode

The default equation is the Landau-Lifshitz-Gilbert (LLG) equation which you have
been using so far. Other equations may be set, e.g. LLB for temperature-dependent
simulations, which will be covered in other tutorials.

There are a number of evaluation methods which you can select. The fixed-step
methods available are: Euler (1st order), trapezoidal Euler (TEuler – 2nd order) and
Runge-Kuta (RK4 - 4th order). The adaptive time-step methods are the adaptive
Heun (AHeun – 2nd order), the multi-step Adams-Bashforth-Moulton (ABM – 2nd
order), Runge-Kutta-Bogacki-Shampine (RK23 – 3rd order with embedded 2nd order
error estimator), Runge-Kutta-Fehlberg (RKF45 – 4th order with embedded 5th order
error estimator), Runge-Kutta-Cash-Karp (RKCK45 – 4th order with embedded 5th
order error estimator), and Runge-Kutta-Dormand-Prince (RKDP54 – 5th order with
embedded 4th order error estimator).

There is also a Steepest Descent solver (SDesc) used for static problems where we
don’t need the magnetization dynamics, but are merely interested in calculating the
ground state (e.g. relaxing a magnetization configuration and hysteresis loops). The
steepest descent solver is only enabled for the LLGStatic equation, which is the LLG
equation with the precession term disabled and damping value set to 1. In this case
the SDesc solver is typically at least an order of magnitude faster in computing the
ground state compared to the other methods.

For most methods the calculated mxh and dmdt stopping condition values are the
maximum |mxh| value (normalized torque) in any given iteration. The exceptions are

53
the Euler, TEuler, and AHeun methods, which are only ever used in practice for
stochastic equations when including a thermal field (more on this later). For this
reason the |mxh| values for these are the mesh averages in any given iteration. This
allows using these methods with an mxh stopping condition for stochastic equations.

As an exercise we will briefly investigate here the stability of the fixed-step methods
as the time step is changed. The time step may be set using:

setdt dt

Here dt is the time value in seconds. For the adaptive time step methods this
command sets the starting time step.

Exercise 6.1

a) Set a 320 × 160 x 20 nm permalloy mesh with a 5 nm cubic cell and relax the
magnetization to |mxh| < 10-5.

b) Set a stage with a magnetic field with components (40 kA/m, 5 kA/m, 0) for 5
ns and save data every 10 ps. Use the RKF45 method. Record the actual
computation time required to complete the simulations. Plot the magnetization
switching dynamics.

c) Repeat the simulation using the RK4 method for fixed time steps of 0.5 ps, 0.7
ps, 0.9 ps and 1.1 ps. Record the actual computation time required to
complete the simulations. Compare the results with the reference results from
the RKF45 method. How do the results change and why?

d) Compare the computation times. Which method is more efficient whilst still
maintaining accuracy?

e) Investigate the computation time required to complete the same problem with
TEuler with a time step of 50fs and Euler with a time step of 30as.

54
Setting shapes

Until now we’ve mostly considered meshes which are filled with magnetic cells. In
general, complex shapes may be generated by masking the mesh using a shape in
an image file. A more advanced method of setting shapes using numerical ovf2 files
is covered in a later tutorial. This is achieved using the following command:

loadmaskfile (zDepth) (directory\)maskfile

In the simplest case the image file defines a 2D shape in black and the void cells in
white – see Figure 6.1 for an example. Instead of using the command you may also
drag and drop the image in the mesh viewer window. The zDepth value defines the
depth the mesh is voided to from top down (if zDepth > 0), or the height the mesh is
voided to from bottom up (if zDepth < 0); this may be used to define 3D shapes with
the maskfile being a grayscale image.

Figure 6.1 – Setting a mesh shape using a mask from a png file

Drag png image file

to mesh viewer to
apply mask

Exercise 6.2

a) Load an ellipse into a 320×160×10 nm mesh. Try not to leave void cells at the
sides. You should use a circle mask as this will be stretched over the defined
mesh rectangle.

55
 b) Obtain hysteresis loops between -50 kA/m and 50 kA/m for the ellipse along
the x axis and along the y axis separately using a cellsize of 5x5x10 nm (2D
simulations). You may use the relaxation condition |mxh| < 10-4 in order to
speed up the exercise. Use the RKF45 evaluation method. How do the two
hysteresis loops compare ?

c) Obtain further hysteresis loops for ellipses with dimensions 260x160x10 nm

and 200x160x10 nm. Compare results for all the simulated ellipses and
explain the changes in the hysteresis loops.

There is a modifier for how shapes are applied to a mesh, accessed using the
individualshape command. By default this flag is Off, and any shape applied to the
mesh is set for all relevant computational quantities. Thus M (magnetisation) is a
computational quantity which may be shaped, but also T (temperature), and 
(electrical conductivity) may be shaped for the relevant solvers (micromagnetics,
heat equation, and transport solver respectively). Note that in order to shape T and 
you need the relevant solvers enabled. If instead you only want to apply the shape to
one of these quantities, or even have different shapes for all of them, you need to
enabled the individualshape flag, then apply the mask. This is useful for example if
you want to include non-magnetic components (e.g. contact leads) in the magnetic
mesh.

You may reset the mesh back to its solid shape using:

resetmesh

This command is also useful to recover the mesh following a wrongly-posed

computation – e.g. if too large a time-step is used the magnetization values will
become NaN (not a number) and must be reset. Other methods to shape a mesh
include setting and deleting rectangles using:

delrect rectangle (meshname)

addrect rectangle (meshname)

56
Tutorial 7 – Magnetocrystalline Anisotropy

In this tutorial you will learn how to use the anisotropy module and simulate
hysteresis loops for different magneto-crystalline anisotropy configurations. You
need to be familiar with all the basic tutorials.

There are two options available for adding magneto-crystalline anisotropy to the
computations: uniaxial or cubic. These are enabled by choosing the aniuni or anicubi
modules from the list displayed using the modules command. The modules are
mutually exclusive, thus enabling one will delete the other one from the list of active
modules.

The strength of the anisotropy is controlled using the K1 and K2 parameters (K1 and
K2 are the anisotropy energy density constants) from the list displayed using the
params command. These are the constants that appear in the anisotropy energy
formulas:

Uniaxial anisotropy:

  
  K1 1  m.e a 2  K 2 1  m.e a 2 
2

Cubic anisotropy:

  K1  2  2   2 2   2 2   K 2  2  2 2 , where
  m.e1 ,   m.e 2 ,   m.e1  e 2 

We also need to define the anisotropy symmetry axes. For uniaxial anisotropy we
only have one symmetry axis and this is set using the ea1 parameter by giving the
Cartesian components of the unit vector ea (e.g. default is 1, 0, 0 for easy axis along
the x-axis). For cubic anisotropy we need two symmetry axes directions, ea1 and
ea2 which should normally be orthogonal.

57
In the following you will investigate the effect of magnetocrystalline anisotropy on
hysteresis loops in circular dots.

Exercise 7.1

a) Set a 160 × 160 × 5 nm permalloy circle with 5 nm cellsize using a mask file.
Set a uniform magnetization along the x direction towards the left (blue state).
Set uniaxial anisotropy with K1 = 10 kJ/m3, K2 = 0 J/m3 and easy axis along x
direction.
b) Simulate hysteresis loops along the x-axis (easy axis), y-axis (hard axis) and
in between along a 45 in-plane direction (remember you will need to use a
polar field sequence for this). You will need to determine appropriate field
sweep ranges so the loops start from a saturated magnetization state.

c) Plot the hysteresis loops using the normalized magnetization (divide by MS

value – the saturation magnetization constant). What are the coercivity and
normalized remanence values? Explain the difference between the loops.

Exercise 7.2

Repeat the simulations in Exercise 7.1, but this time set cubic anisotropy with K1 =
20 kJ/m3 and K2 = 0 J/m3, with two perpendicular easy axes in the plane (e.g. x-axis
and y-axis).

Plot the resulting hysteresis loops and compare them with the previous results.

Hints:

For the 45 direction you will need to project the magnetization along the applied
field direction. You can do this by taking the dot product of M with the applied field
direction unit vector:

58
 ˆM
M H  h.

You can do this using console data processing with the command:

dp_dotprod dp_vector ux uy uz dp_out

Here dp_vector is the dp array index such that the dp arrays dp_vector, dp_vector +
1, dp_vector + 2 hold the x, y, and z components of the magnetization, (ux, uy, uz)
are the components of a vector (dot product taken with this vector), and dp_out is the
dp array where the output is placed.

Finally, you can normalize the magnetization using the dp_div command where you
will need to divide by MS. To see the value of MS you can look it up in the list
displayed using the params command.

Remember you will first need to load into dp arrays the appropriate columns for the
saved hysteresis loop data file using the dp_load command. You can save the
contents of dp arrays after processing data using the dp_save command. As always
you can find more details about a command by preceding it with the ? symbol, e.g.
?dp_load.

59
Tutorial 8 – Anisotropic Magneto-Resistance

In this tutorial you will learn how to simulate magneto-resistance loops and calculate
charge current densities using the transport module.

Transport Module Basics

The transport module is a complex spin and charge current solver (electron
transport), allowing for a number of physical effects to be included in the
magnetization dynamics problem, including Zhang-Li spin-transfer torques based on
calculated charge currents, spin torques based on computed spin accumulations in
multilayers, direct and inverse spin Hall effects (SHE and ISHE), spin pumping
torques, anisotropic magneto-resistance (AMR), current-perpendicular-to-plane giant
magneto-resistance (CPP-GMR), Oersted fields and Joule heating.

Here we will look at how a simple charge current density may be computed and AMR
included in the simulation.

You will first need to enable the transport module from the list displayed using the
modules command for the meshes where you want a charge current density to be
computed. In the simplest case the computation is reduced to obtaining J, the
charge current density, using Ohm’s law : J = E, where  is the electrical
conductivity and E = -V is the electrical field with V being the electrical potential. If
 is constant this reduces to a Laplace equation for V.

The electrical conductivity, potential and charge current density are available as
display outputs under the display command (elC, V and Jc).

The base electrical conductivity value may be changed by editing the elC mesh
parameter displayed using the params command.

Similarly AMR may be enabled by editing the amr mesh parameter (0% by default
which disables it). A typical value for permalloy is 2%. Enabling AMR results in a

60
non-uniform electrical conductivity and now the equation for V becomes a Poisson
equation.

Before starting a computation you will need to define at least 2 electrodes – these
set Dirichlet boundary conditions for V (fixed potential values) in the Laplace/Poisson
solvers. The most common electrode configuration is to define two electrodes at the
x-axis ends of the mesh (so in the y-z plane). You can do this using the command:

setdefaultelectrodes

To see which electrodes have been defined use the command:

electrodes

You will see two electrode rectangles. The electrode rectangles are in absolute
values, so not relative to any particular mesh. You can add new electrodes but they
must always be placed at the edges of a mesh rectangle – when initializing the
simulation Dirichlet boundary conditions will be flagged for the boundary cells of the
mesh intersecting the electrode rectangle. Each electrode has a fixed potential which
may be edited. Exactly one of the electrodes has to be designated as the ground –
this is the electrode where the outgoing total electrical current is calculated.

You can edit the individual electrode potential values, however a more common
scenario is the set a single electrical potential drop from the ground to the other
electrodes using:

setpotential potential

This sets a single inversely-symmetrical potential drop (i.e. +potential/2 to

-potential/2). The inversely-symmetrical potential drop minimizes floating point errors
(as opposed to setting a potential drop of potential to 0).

Simulations may use the constant-voltage or the constant-current mode (the

interactive object displayed when using the electrodes command may be toggled
61
between these two states). Normally you would use the constant-voltage mode; with
constant-current the electrode potentials are adjusted during the simulation to
maintain a constant current (which may be set using the setcurrent command). In
this tutorial we will be using the constant-voltage mode.

Exercise 8.1

Set a permalloy mesh with dimensions 320x320x10 nm and mask it using a circle
shape (in the image file make sure to not leave any white spaces at the left and right
sides).

Enable the transport module, set the default electrode configuration and a potential
of 10 mV.

In the data box display the average current density (Jc), set potential (V), total
current (I) and resistance (R). In the mesh display the current density. Run the
simulation. The calculated current density should look similar to that in Figure 8.1.

Figure 8.1 – Computed charge current density for Exercise 8.1

62
For more advanced simulations you can add electrodes using the addelectrode
command. Electodes may be deleted using the delelectrode command (or more
simply by right-clicking on an existing electrode in the list displayed by the
electrodes command. All electrodes may be deleted by using the clearelectrodes
command.

Further info:
When displaying the meshes (use the mesh command) you will now notice a value
for the electric cell. This is the discretization cellsize used by the transport solvers.
Normally this should be equal to the magnetic cellsize (default setting) but can be
controlled separately for more advanced simulations (decrease computation time or
increase computation accuracy as required). Multiple meshes with transport modules
enabled may be configured. If the meshes are touching, composite media boundary
conditions will automatically be inserted in the computation, however we still typically
require 2 electrodes for a well-posed problem.

Exercise 8.2

Set a permalloy rectangle with dimensions 300x100x20 nm. Calculate the current
density for the default electrodes setting by using a potential drop of 1 V.

What is the computed sample resistance and does it agree with that predicted by the
formula:
l
R ,
A

where l is the length, A the cross-sectional area and  is the resistivity.

What is the total current, and does it agree with the expected value for a 1 V
potential drop?

63
What is the average current density and does it agree with the expected value for the
total current?

Further info:
For advanced simulations the accuracy of the transport solver may be controlled by
using the command:

tsolverconfig

This command displays the set convergence error (a value around 10-6) is normally a
good compromise between accuracy and computational speed. In this case the
Laplace/Poisson solvers stop iterating when the maximum change in V from one
iteration to another, normalized to the set potential drop, drops below this set
convergence value. You can display the ts_iter (current number of transport solver
iterations) and ts_err (current transport solver error) data in the data box. If the
convergence error threshold is too low the transport solver will take a large number
of iterations during computations – if AMR, GMR, ISHE or temperature-dependent
transport parameters are enabled the transport solver must update after every
magnetization and/or heat solver time step.

The transport solver may be used to calculate magneto-resistance loops when an

AMR (anisotropic magneto-resistance) value is set. Since this is a static problem it is
best to set the LLGStatic equation with the SDesc solver. When using the SDesc
solver the stopping condition (mxh or dmdt) should be much lower than usual,
typically at least 10-6 or even lower. You should also set the static transport solver
flag to On (see tsolverconfig command output). Setting this flag to On will only
iterate the transport solver at the end of a schedule step, thus resulting in faster
computation. In this case you should also increase the iterations timeout to at least
5000 or more to ensure the transport solver converge threshold is met.

64
For problems where the current density is uniform a transport solver converge
threshold of 10-6 is sufficient, however this may have to be decreased to 10 -8 or even
10-9 if the current density is highly non-uniform.

In the following problem you should use both the static LLG equation and static
transport solver as explained above.

Exercise 8.3

Here you will obtain longitudinal and transverse magneto-resistance loops.

a) Set a permalloy rectangle with dimensions 160x80x10 nm. You can use a 2D
simulation with magnetic cellsize 5x5x10 nm, but the transport solver should
be left with cubic electric cellsize of 5 nm.

Sweep the field from -100 kA/m to +100 kA/m strength and back, using a field
step of at most 1 kA/m, along a nearly longitudinal direction (use 5 from the
x-axis). You should use a Hpolar_seq sequence, and an mxh stopping
condition of 10-7.

Set the transport solver with default electrodes, a non-zero potential drop (e.g.
1 mV), and amr = 2% (see params).

In the output data make sure to save the applied field and sample resistance
every step. You should also save the transport solver error to check the
converge threshold has been met at the end of each step (ts_err).

Run the simulation and plot the obtained MR loop. Explain your results.

65
 b) Repeat the simulation along the in-plane nearly transverse direction (use 85
from the x-axis) by sweeping the field between -100 kA/m and +100 kA/m
strength with a step of 1 kA/m.

Estimate the AMR percentage from the obtained H vs R loops from parts a)
and b) – does it agree with the set 2% value?

Figure 8.2 – Longitudinal and transverse MR loops due to AMR as calculated in

Exercise 8.3.

66
Tutorial 9 – Scripting using Python

In this tutorial you will learn how to use scripts to automate multiple simulations.
Boris can communicate with an external program via network sockets, so both local
and remote script-based control is possible. For this purpose a Python module
(NetSocks) is provided, allowing use of Python scripts to communicate with Boris.
This is contained in the NetSocks.py file and requires Python 3.7.

The scripts work by sending console commands, with syntax identical to that you
would use when typing commands directly in the console. Some console commands
also return values, which can be read by Python scripts. You can find out if a console
command is set to return values by looking at the USAGE help for it. For example
type the following in the console:

dp_coercivity

In the USAGE help you can see this command will return the calculated coercivity
value (see the description “<script return values: firstcoercivityvalue>”).

NetSocks Python Commands Usage

Look through the examples in Tutorial 0. To create a new simulation script:

1. Place the NetSocks.py file in the same directory as your Python simulation
script file and import NSClient from NetSocks.

2. Next a NSClient object, ns, must be created as:

ns = NSClient('localhost')

If running the script remotely the localhost entry needs to be replaced by the
IP address (port 1542 is used, which must not be blocked in the firewall) of
the computer running the Boris executable.

67
 Now all communication is done through methods in the ns object.

All commands can be sent through the ns object as ns.command(parameters…), for

example ns.setfield(90, 0) sets a magnetic field along the x axis.

You can obtain returned parameters as data = ns.command(parameters…), for

example Mav = ns.showdata(‘<M>’) returns the average magnetisation as a list with
3 elements.

In many cases the script must wait for a simulation to finish before proceeding. This
is achieved by using the Run function:

 ws.Run()

This is a blocking function call, which sends the run command, then waits for the
simulation to finish by listening to messages from the running program.

NSClient also has a useful method for writing data to a file:

 ns.SaveDataToFile(‘Results.txt’, [data1, data2, …])

This command appends a new line in the Results.txt file (in the script directory)
which contains two numbers: the x and y components of magnetization.

Look through the examples in Tutorial 0 and run them.

Examples of other scripted simulations will be contained in further tutorials.

68
Tutorial 10 – Current-Induced Domain Wall Movement

In this tutorial you will learn how to obtain current-induced domain wall velocity
curves. First we will simulate these using only console commands, including
processing of output data, then we will use a Python script to more accurately
determine the domain wall velocity.

The simplest available method for enabling spin torques is to use the Zhang-Li spin-
transfer-torque (STT) formulation. In this formulation the calculated charge current
density is used to calculate the following spin torques on the magnetization (included
as additive terms in the normalised LLG equation):

TSTT  u. m  m  u. m

where
P B 1
u  JC
MS e 1  2

To enable this use the ode command and select the LLG-STT equation with the RK4
evaluation method. You will also need to enable the transport module (use the
modules command and select the transport module). As before you need to set two
electrodes (setdefaultelectrodes) and enable the moving mesh algorithm
(preparemovingmesh).

In the above equation we have two new parameters: P, the charge current spin
polarisation, and , the STT non-adiabaticity parameter. These can be edited by
using the params command and double-clicking on the respective interactive
console objects; the default values are set for permalloy.

Exercise 10.1

a) Set a 320 × 80 × 10 nm permalloy rectangle with 5x5x5 nm cellsize (3D

problem). Enable the moving mesh algorithm and let the domain wall relax in

69
 zero field. Enable the transport module, set the default electrode configuration
and enable the LLG-STT equation with the RK4 evaluation method.

b) Set a simulation schedule to vary the current density from 10 11 A/m2 up to

1012 A/m2 in 10 steps, saving the domain wall shift (dwshift), current (I),
voltage (V), and charge current density (Jc) output data. Each current density
value should be maintained for 5 ns with a data saving schedule every 10 ps.

Hint: You should use the V_seq simulation schedule stage. This sets a sequence of
voltage values with given start and stop values in a number of steps. Since we have
a simple geometry you can calculate the required voltage values using the formula:
l JC
V


where l is the distance between electrodes (length of the magnetic mesh) and  is
the electrical conductivity (see the set value using the params command). Solution:
use V_seq with -4.57mV; -45.7mV; 10.

Note: You can also vary the current density using the I_seq schedule stage – this
defines a sequence of current values. Setting current values directly enables the
constant current mode, i.e. the voltage drop between electrodes is adjusted during
the simulation to keep the current constant (the opposite of this is the constant
voltage mode).
You can see the current settings using the electrodes command.
You can also set individual values in the console using the setpotential or
setcurrent commands.

c) Obtain and plot the domain wall velocity, v, using the method introduced in
Tutorial 5 with dp_linreg, as a function of current density. Convert the current
density to spin drift velocity using the formula:

P B 1
u  JC
MS e 1  2

70
 Verify that the following formula holds:

v 

u 

Domain wall velocity curves may also be obtained using a Python script, allowing
more accurate determination of the velocity. The problem with the dp_linreg
method, it also includes in the regression domain wall displacement points at the
start of each step. Since the domain wall requires some time to reach a steady state
velocity (the acceleration is not zero at the start of each step) these initial
displacement values should ideally be discarded.
With a Python script you can set a single voltage stage without saving any data (e.g.
for 3 ns); this is followed by another voltage stage (e.g. again for 3 ns) during which
data is saved, then a linear regression is performed to extract the velocity value for
the set voltage value. The script then proceeds to set the next voltage value and so
on.

Exercise 10.2

Repeat the simulation in Exercise 10.1 but using a Python script to control the
simulation flow and data output.
(Solution: see the attached Python script)
Note, in general you may need to adjust the stage duration times (1 – to achieve
steady state motion and 2 – to generate enough data to obtain a representative
velocity value using linear regression) for best results, but the suggested 3 ns, 3 ns
breakdown will be sufficient for this exercise.

When setting up the Python script you will need to use the appropriate commands to
edit the stage stopping and saving conditions, as well as the stage value. These
commands are:
edistagevalue
editstagestop
editdatasave

71
Tutorial 11 – Oersted Fields

In addition to spin-transfer torques, electrical currents may also interact with

magnetization via the generated Oersted fields. In this tutorial we will set-up a simple
bilayer mesh structure, consisting of a magnetic wire and a non-magnetic metallic
capping layer, then repeat the domain wall velocity simulations from the previous
tutorial but also taking into account the generated Oersted field. Since the bilayer
structure has a broken mirror symmetry in the z direction we might expect the
domain wall speed to be different depending on the current direction.

To enable the Oersted field you need to enable the Oersted module (use the
modules command). The Oersted module is an electric super-mesh module, i.e. it is
calculated on the electric super-mesh with a separately controlled cell-size. After
enabling the Oersted and transport modules bring up the configured meshes using
the mesh command. You should now notice the electric super-mesh rectangle with
its cellsize; the electric super-mesh is simply the smallest rectangle containing all
meshes with the transport module enabled.

To set a non-magnetic metallic capping layer you will need to add a new mesh, in
particular an electrical conductor mesh. This is done using the command:

addconductor name rectangle

72
Exercise 11.1

Set-up a simulation space as for Exercise 10.1. Add a metallic capping layer on top
of the magnetic layer with a 5 nm thickness and enable its transport module
(solution: use addconductor cap 0 0 10nm 320nm 80nm 15nm).

Set the electrodes to contact both meshes (use setdefaultelectrodes after both
meshes had their transport module enabled).

Enable the Oersted field module and set a potential of 10mV (setpotential 10mV).

Compute a single iteration (computefields) and display the Oersted field (use the
display command and select the Oersted display option on the super-mesh display
line).

Figure 11.1 – Computed Oersted field for Exercise 11.1.

73
Exercise 11.2

Continuing from Exercise 11.1, use a Python script to simulate the domain wall
velocity for both positive and negative currents in the magnitude range 1011 A/m2 to
1012 A/m2 in 10 steps. Plot the two ve
locity curves and compare them to the expected v = (/)u relation, as well as the
simulations without an Oersted field. (Solution: see the Python script in the tutorial
resources).

Note: when plotting v vs u you need to set the sign of u to be in the direction of
electron drift, i.e. opposite sign to that of the charge current density.

In the above exercise the default electrical conductivity value of the capping layer is
the same as for permalloy (this can be edited using the params command). In this
case the current densities are the same in both layers. If you want to save output
data for a particular mesh (e.g. Jc, the charge current density, which is mesh
dependent) then you must focus on the required mesh first before adding that
particular data to the output list.

You can focus on a particular mesh by clicking on the mesh name (bring up the list
of meshes with mesh then click on a mesh name). Alternatively you can double click
in the mesh graphical viewer on the required mesh.

Results from Exercise 11.2 are shown in Figure 11.2. The data obtained in Figure
11.2 could be improved further. The problem with using linear regression directly on
the raw dwshift data, it contains steps due to the mesh discretisation. This is
particularly problematic at low domain wall velocities where only a few steps are
contained in the raw data, which can make the extracted velocity inaccurate.

74
Figure 11.2 – Simulation results from Exercise 11.2 showing a) domain wall velocity
plotted against spin drift velocity for cases with and without Oersted fields and also
compared with the analytical formula, and b) domain wall speed difference plotted
against spin drift speed for cases with and without Oersted fields.

One possibility is to collect data for a longer time, but this is inefficient. Another
possibility is to assume the domain wall displacement is linear with time (in this
exercise this is a valid assumption). You can then either get rid of the repeated
points before using linear regression, or replace the repeated points using linear
interpolation. There is a built-in command for this, and is covered in a later tutorial on
skyrmion movement (dp_replacerepeats).

75
Tutorial 12 – Surface Exchange, Multi-Layered Demagnetization and
CUDA

Surface Exchange

Using the surfexchange module, two or more ferromagnetic meshes can be surface
exchange coupled, allowing simulations of magnetic multilayers with RKKY
interaction. The strength of the surface exchange coupling is controlled using the J1
and J2 material parameters: negative values result in anti-ferromagnetic coupling,
positive values in ferromagnetic coupling. The J1 parameter controls the strength of
bilinear surface exchange, and J2 controls the strength of biquadratic surface
exchange.

Type params and have a look at J1 and J2. For two meshes in surface exchange
coupling, it is the top mesh J1 and J2 values that are used. This allows setting
different coupling strength and types for the bottom and top of a mesh in a multi-
layered structure. Boris allows surface exchange coupling only for xy planes, thus a
multi-layered structure should be designed with the layers stacking along the z
direction. Two ferromagnetic meshes will be surface exchange coupled if they both
have the surfexchange module enabled and there’s no other ferromagnetic mesh
with the surfexchange module enabled in between them along the z direction. The
coupling will only be calculated for cells which overlap in the xy plane.

Multi-Layered Demagnetization

To add another ferromagnetic mesh use the addmesh command. The demag
module for each mesh only calculates the demagnetizing field for that ferromagnetic
mesh alone. When 2 or more ferromagnetic meshes are being used, if you want to
compute the overall demagnetizing field you need to use the supermesh sdemag
demagnetizing field module. In this case the individual demag modules are disabled
and the overall demagnetizing field is computed for the collection of individual
ferromagnetic meshes, including all stray field contributions. There are two ways to

76
do this. The default method is called multi-layered convolution, and you can see the
settings for this by using the command:

multiconvolution

This is an exact method of computing demagnetizing fields for a collection of

computational meshes, which is able to handle arbitrary spacing and relative
positioning between the layers without sacrificing accuracy or computational
performance – see S. Lepadatu, Journal of Applied Physics 126, 103903 (2019) for
details. In many cases you can rely on the default settings for multi-layered
convolution, but for more advanced control you should read the information below.

Further info:
You can force the demagnetizing field to be computed in a 2D approximation in each
mesh by clicking on the respective interactive console object (see output of the
multiconvolution command, Force 2D : 2D meshes). This allows each mesh to
have an arbitrary thickness, and is appropriate if the meshes are thin enough for the
2D approximation to hold. This option is turned off by default (Force 2D : Off),
resulting in a 2D or 3D convolution in each mesh depending on their cellsizes. If the
2D approximation for each mesh is not appropriate you should use this option as
long as the z cellsizes are the same in all meshes. If the z cellsizes differ the
computation with Force 2D : Off may not be accurate, and in this case you should
use the 2D layering option (Force 2D : 2D layered). This is similar to the 2D meshes
option, however instead of forcing each mesh to be 2D, it is instead split into multiple
layers, with each layer thickness set by the z cellsize in each respective mesh. This
option is thus the most general case and allows exact computation of demagnetizing
fields without any approximations, however it is in general more computationally
expensive and should only be selected if the other 2 options are not appropriate.

For multiple computational meshes with unequal sizes, the algorithm works by first
transferring the magnetization values to scratch spaces with a common
discretization. You can specify what this discretization should be, but by default it is
calculated for you (see output of the multiconvolution command).

77
Another method of calculating demagnetizing fields for a collection of computational
meshes is to use the so-called supermesh demagnetization. This is achieved by
disabling the multi-layered convolution algorithm (see output of the
multiconvolution command). This method calculates the demagnetizing field on the
ferromagnetic supermesh by transferring magnetization and demagnetizing field
values to and from the ferromagnetic supermesh using a local averaging smoother.
Remember the ferromagnetic supermesh is the smallest rectangle containing all the
ferromagnetic meshes and can be viewed using the mesh command. Computations
on the ferromagnetic supermesh are done using its independent cellsize as can be
seen in the console output of the mesh command. This cellsize will need to be
carefully determined in each case to ensure accuracy of results. As a rule you should
set the cellsize to be the minimum value out of the individual mesh cellsize values
required to compute the demagnetizing field accurately separately. In most cases of
interest, if full accuracy is required, this method is much slower than multi-layered
convolution. It is also far less flexible, as it cannot accurately handle spacing
between layers which cannot be exactly discretized.

Another use for the sdemag module without multi-layered convolution, is to calculate
the demagnetizing field in an individual mesh with a different cellsize to that used for
the exchange interaction. The exchange interaction typically requires a smaller
cellsize to ensure accuracy, thus this method can be used to improve computational
speed for larger simulations.

Starting from the default state, add another ferromagnetic mesh with dimensions of
80 nm × 80 nm × 20 nm, separated from the first mesh by 2 nm along the z direction.
Enable the surfexchange module for both meshes, as well as the sdemag module.
Now run the simulation and observe the result – see Figure 12.1.

78
Figure 12.1 – Anti-ferromagnetic surface exchange coupled ferromagnetic meshes

You can also add a metallic spacer layer in between, using the addconductor
command, however this will not affect magnetic computations directly; it will be
needed however if charge or spin transport computations are also enabled.

Exercise 12.1

a) Set-up a synthetic ferrimagnetic (SyF) Ni80Fe20 bilayer with elliptical shape

of 320 nm × 160 nm, thickness values of 20 nm and 10 nm respectively, and
with a separation of 2 nm between layers. Simulate a hysteresis loop for this
SyF structure along 1 to the x-axis direction, plotting the average
magnetization for the entire bilayer against field (you will need to calculate this
from the magnetization saved for the two layers separately – see notes
below).

You should use the SDesc with LLGStatic solver and set a low mxh threshold
value, in this case it is recommended to set it to 10-7 (use ode command). The
field step should not be greater than 1 kA/m.

b) Repeat the same exercise but this time set-up a synthetic antiferromagnetic
bilayer (SAF) with the same overall thickness as above – i.e. 15nm thick
layers with 2 nm spacing.

When adding data to the output list, you can select the mesh for which it applies, if
applicable, e.g. magnetization output. You can do this either by adding data when

79
the required mesh is in focus, or editing that data entry later to change the applicable
mesh name. For the exercise above you will need to add to the output the
magnetization for both meshes as two separate entries (use data command and
follow instructions therein).

You will need to apply the field for the hysteresis loop to both meshes, not just the
first permalloy mesh. Use the stages command and add a Hpolar_seq stage. You
will see the field is set to be applied only to the current mesh in focus, e.g.
Hpolar_seq <permalloy>. Instead, you will need to edit this by double-clicking on the
added stage entry, and changing the name from permalloy to supermesh. After this
the entry should read Hpolar_seq <supermesh>. The field sequence will now be
applied to both ferromagnetic meshes.

Figure 12.2 – Hysteresis loops obtained for the SyF and SAF bilayers in Exercise
12.1

When working with multiple ferromagnetic meshes, all the commands that affect
changes in a ferromagnetic mesh have an optional parameter which specifies which
mesh to use. If only one ferromagnetic mesh is created the mesh name doesn’t need
to be specified explicitly. If multiple meshes are used, unless the name is specified

80
the settings are applied either to the current mesh in focus, or to the supermesh,
depending on the command.

For example the setangle and setfield commands will make changes to all
ferromagnetic meshes unless a specific name is specified – see the help for these
commands (?setangle, ?setfield). On the other hand the loadmaskfile command
(remember you can just drag a .png file to the mesh viewer instead of typing this
command) only applies the shape to the current mesh in focus.

CUDA Computations

If you have a CUDA-enabled graphics card you can enable GPU computations using
the cuda command:

cuda 1

If CUDA computations are not available for your computer, typing the cuda
command will show an N/A status. You can also see how much CPU and GPU-
addressable memory you have by using the memory command.

When using CUDA computations, for optimum efficiency you will want to limit the
display update frequency (use the iterupdate command). Boris is designed to limit
memory transfers between GPU and CPU-addressable memory to an absolute
minimum, as this is a critical bottleneck in performance. To display mesh data in the
viewer, an average display data is computed on the GPU, then transferred to CPU-
addressable memory so it can be used by graphics routines. This transfer can slow
down computations, especially if the display viewing coarseness is small (remember
you can use the mouse wheel to change viewing coarseness).

If the simulation is very large you should be careful about setting too small a viewing
coarseness, as this will make the program slow and less responsive when working
with the console.

81
Tutorial 13 – Dzyaloshinskii-Moriya Exchange

Dzyaloshinskii-Moriya interaction (DMI) may be included in the simulation by

enabling the DMExchange, or the iDMExchange module. The former is used for bulk
DMI, whilst the latter is used for interfacial DMI. The strength of the DMI interaction is
controlled using the D material parameter (use params command). Néel skyrmions
may be generated in the xy plane using the skyrmion command:

skyrmion core chirality diameter position

In the above command the core parameter sets the z direction of the skyrmion core
(-1 or 1), the chirality parameter sets the radial direction rotation (-1 for away from
core, 1 for towards core). The diameter and position may be specified using metric
units, with the position requiring 2 components. Figure 13.1 shows examples of
relaxed Néel (iDMExchange) and Bloch (DMExchange) skyrmions for both D > 0 and
D < 0.

Figure 13.1 – a) Bloch skyrmion for D < 0, b) Bloch skyrmion for D > 0, c) Néel
skyrmion for D < 0, and d) Néel skyrmion for D > 0.

82
For the following exercises, when computing a relaxed magnetization state you
should use the SDesc with LLGStatic solver, remembering to set a low mxh
convergence value (at least 10-6).

Exercise 13.1

a) Obtain relaxed Néel skyrmions for both D > 0 and D < 0 in an ultrathin (1 nm)
Co layer with perpendicular magnetization. You will need to use the
iDMExchange module. Use material parameters as Ms = 600 kA/m, A = 10
pJ/m, |D| = 1.5 mJ/m2, K1 = 380 kJ/m3 for uniaxial anisotropy with easy axis
along z direction. To reduce the skyrmion diameter apply an out-of-plane
magnetic field opposing the skyrmion core, e.g. 15 kA/m along the 0, 0 (polar
coordinates) direction.

b) Obtain Bloch skyrmions for both D > 0 and D < 0. You may use the same
parameters as above, but this time set a thickness of 10 nm and use the
DMExchange module. You will need to set a larger out-of-plane magnetic field
to control the skyrmion diameter.

c) For part a) compute the skyrmion diameter and topological charge.

To complete Exercise 13.1 c), you will need to extract a profile along the skyrmion
diameter, then fit it using an analytical model for a skyrmion. Boris provides this
functionality through the dp_fitskyrmion command. The z component of M is fitted
using the following formula (see X.S. Wang et al., Commun. Phys. 1, 31 (2018)):

  sinh R / w   D
M Z ( x)  M S cos 2 arctan  , w  , K  K u   0 M S2 / 2
  sinh  x  x 0 ) / w  
 4 K

Here MS (saturation magnetization), R (skyrmion radius), x0 (skyrmion center

position), and w are used as fitting parameters, and in particular the Ms and w values
after the fit should match the expected values from the material parameters set.

83
Thus to obtain the skyrmion radius first use dp_getprofile to extract a profile through
the center of the skyrmion, then use dp_fitskyrmion on the z component of M. The
fitting works for both skyrmion topological charges, but the fitted value of MS will
change sign depending on the sign of the topological charge.

To calculate the topological charge, there is a built-in command, dp_topocharge.

This command solves the following formula over the xy plane S:

1  dm dm 
4 S  dx dy 
Q m.  dxdy

Exercise 13.2

For the same parameters as in Exercise 13.1, compute the skyrmion diameter as a
function of out-of-plane field strength from 2 kA/m to 20 kA/m in 1 kA/m steps. You
should set-up a Python script to automate this simulation.

Figure 13.2 – Skyrmion diameter as a function of out-of-plane field strength obtained

in Exercise 13.2.

84
Tutorial 14 – Simulations with non-zero temperature

Landau-Lifshitz-Bloch equation and Curie temperature

Non-zero temperature simulations may be performed by using the Landau-Lifshitz-

Bloch (LLB) equation – use the ode command then select the appropriate equation
to solve.
With a non-zero temperature the magnetization length (saturation magnetization) is
no longer a constant, but depends on the applied field strength. This is modelled via
a longitudinal susceptibility included in the LLB equation. Instead, we talk about the
equilibrium magnetization, which is the stable magnetization length at a given field.
Thus at zero temperature the equilibrium magnetization coincides with the saturation
magnetization. With a non-zero temperature the equilibrium magnetization gradually
decreases, reaching zero at the Curie temperature. Other parameters which change
with temperature include the exchange stiffness and magnetization damping. With a
non-zero temperature the damping is now divided into two terms: transverse
damping (coincides with the Gilbert damping at zero temperature) and longitudinal
damping. For further information these articles can be used as a starting point: S.
Lepadatu, Journal of Applied Physics 120, 163908 (2016) and S. Lepadatu & M.M.
Vopson, Materials 10, 991 (2017).

In the simplest case the temperature inside the mesh is uniform, and is controlled
using the temperature command, which sets the mesh base temperature:

temperature value (meshname)

When enabling the LLB equation you will also need to set appropriate temperature
dependences for some material parameters, including Ms, damping, A, and susrel
(the relative longitudinal susceptibility). The longitudinal damping used in the LLB
equation is not available as a separate material parameter, but is automatically
calculated based on the transverse damping parameter (damping). Default
temperature dependences for these parameters may be generated based on the
Curie temperature of the material – for details see S. Lepadatu, Journal of Applied
85
Physics 120, 163908 (2016). You can do this using the curietemperature
command:

curietemperature value (meshname)

Setting material parameters temperature dependences

Exercise 14.1

Set a Curie temperature of 870 K (appropriate for Ni80Fe20) and obtain plots of the
temperature dependences of the Ms, damping, A, and susrel material parameters
(see below).

Almost all material parameters available in Boris can be assigned a temperature

dependence. This is achieved by specifying a scaling law, t. The value of a
parameter at a temperature T is then obtained as value_at_T_K = value_at_0_K ×
t(T) – any computational routine in Boris for a which a parameter is used, obtains an
updated value in this way where appropriate, where T is either the base temperature
(uniform temperature mode) or the local temperature (non-uniform temperature
mode). To see the currently set temperature dependences use the paramstemp
command. You can set a temperature dependence by supplying a text equation,
where T is the temperature value (see console output for paramstemp), or by
loading an array using the setparamtemparray command. Further details on using
text equations are given in a dedicated tutorial. Once a temperature dependence
array has been set (e.g. after using the curietemperature command), you can load
the set temperature dependence into an internal data processing array using the
dp_dumptdep command:

dp_dumptdep meshname paramname max_temperature dp_index

For example to see the set temperature dependence for Ms use the following:

dp_dumptdep permalloy Ms 870 0

86
dp_save Ms_scaling 0
In the file Ms_scaling.txt (saved under the current working directory – see data
command) you will see a single column with the scaling coefficients. These are
saved in increments of 1 K, from 0 K up to 870 K. Internally the scaling coefficients
are obtained from the user loaded array at 1 K increments, irrespective of how the
user specified the temperature dependence – missing temperature points are filled in
using interpolation. During computations the scaling coefficients are obtained by
interpolating the nearest 2 temperature scaling points. To reset all parameters
temperature dependences you can use the clearparamstemp command.

Field dependence of material parameters temperature dependences

With non-zero temperature simulations, the equilibrium magnetization also depends

on the strength of the applied magnetic field. This dependence is enabled by setting
the strength of the net atomic moment of the material, specified in Boris as multiples
of the Bohr magneton. This is done using the command:

atomicmoment (value)

If this value is not zero, whenever the applied magnetic field changes, the
temperature dependences of all the parameters affected by the Curie temperature
setting (see above) are recalculated. Moreover, the longitudinal susceptibility is
directly proportional to this value so must be set correctly whenever the LLB
equation is used.

Non-zero temperature simulations

Exercise 14.2

Simulate the hysteresis loops in a 160 × 160 × 5 nm permalloy circle at zero

temperature (using the LLG equation), as well as at room temperature (297 K, using
the LLB equation) and compare the two loops. With the LLB equation the time step
for numerical stability is usually lower – you might need to investigate this.

87
Tutorial 15 – Thermal Fields

When non-zero temperature modelling is considered, an additional effect that can be

included is lattice thermal agitation. This gives rise to fluctuations in magnetic
moments, and may be modelled by introducing appropriate stochastic fields and
torques. In Boris thermal fields may be enabled by selecting a stochastic
magnetization dynamics equation, e.g. sLLB – use the ode command then select the
appropriate equation to solve. For further information see S. Lepadatu & M.M.
Vopson, Materials 10, 991 (2017).

When solving stochastic equations, the choice of available ODE evaluation methods
is more limited since they must be able to handle the stochasticity introduced.
Currently the best fixed time-step method available in Boris is the trapezoidal Euler
(TEuler) evaluation method, also known as Heun’s method. Since this method is a
fixed time step method you will need to investigate the time step required for
numerical stability. You can use the default time step as a starting point.

There is an adaptive time-step version of this called AHeun which you can also use.

Exercise 15.1

Simulate out-of-plane hysteresis loops at room temperature in a 256 nm × 256 nm

Co rectangle with perpendicular magnetization, with 4 nm thickness, and cubic 4 nm
cellsize. Use material parameters as Ms = 600 kA/m, A = 10 pJ/m, and K1 = 380
kJ/m3 for uniaxial anisotropy with easy axis along z direction. You should simulate
hysteresis loops with and without thermal fields for comparison.

88
Tutorial 16 – Heat Flow Solver and Joule Heating

Heat equation

Non-uniform temperature simulations may be enabled by selecting the heat module.

Any mesh with this module enabled will solve the heat equation as a function of time.
If any two meshes with the heat module enabled are in contact, then heat flow
across the interface (also referred to as a composite media boundary) is
automatically calculated based on the continuity of heat flux and temperature
perpendicular to the composite media boundary.

There is a special type of mesh, referred to as an insulator mesh in Boris, which can
be used to model substrates. You can add an insulator mesh using:

addinsulator name rectangle

When the heat module is enabled, the thermal cell discretisation cellsize becomes
available in the mesh descriptions (use the mesh command). This can be controlled
independently of the magnetic and electric cellsize (if enabled), and can also be set
independently of other cellsize values in other meshes.

The heat equation time step may be set using:

setheatdt value

This value shouldn’t be larger than the magnetization dynamics equation time step
(setdt), since during computations the heat equation time is incremented only up to
the current magnetization equation time (the global time, or total time – see the time
output data). If this value is lower, the heat equation will be iterated multiple times
until it catches up to the magnetization equation time.

The mesh temperature may be set as before using the temperature command. This
sets a uniform mesh temperature as a starting point, but depending on the simulation

89
configuration the mesh temperature can change. This is true particularly if the mesh
ambient temperature is different. For the heat equation, boundary conditions for cells
not at a composite media boundary are set based on Newton’s law of cooling – i.e.
Robin boundary conditions are used. These require an ambient temperature (the
surrounding temperature) and a heat transfer coefficient (the Robin coefficient). To
adjust these values you may use the ambient command, then double click on the
respective interactive objects to modify their values. Note, when the temperature
command is used, setting a mesh temperature automatically sets the ambient
temperature to the same value too. You may also choose to have thermally
insulating boundary conditions by selecting the appropriate options displayed by the
ambient command.

Parameters for heat transport

A few parameters are used to specify the thermal properties of the material, in
particular the thermK (thermal conductivity) and the shc (specific heat capacity)
material parameters – see these by using the params command. Additionally the
density (mass density) parameter also enters the heat equation.

Note, all material parameters with a temperature dependence enabled will now also
vary non-uniformly throughout the mesh (if heat module enabled), taking on the
value set by the local cell temperature value.

You may obtain the mesh average temperature through the <Temp> output data –
use data command. The mesh temperature may also be displayed (Temp) – use the
display command.

Joule heating

If the transport module is also enabled in the same mesh as the heat module, Joule
heating is taken into account. This results in a heat source term in the heat equation
due to the charge current density, J as:

90
 T (r, t ) J2
Cρ  .KT (r, t ) 
t 

In the next exercise you will investigate the effect of a voltage pulse on a Ni 80Fe20
nanowire placed on a SiO2 substrate, similar to the work in S. Lepadatu, Journal of
Applied Physics 120, 163908 (2016). To model a very long wire on a substrate (say
the wire is oriented along the x-axis) you should set the x-axis ends of both the
magnetic wire and substrate as insulating since in this case the heat flux is oriented
only along the y and z directions. Similar considerations apply to the substrate if you
want it to be effectively infinite in the x-y plane and depth – set insulating boundary
conditions in the required directions. Note in this latter case the modelled substrate
must still be large enough for the temperature evolution to be correct for the required
duration – for details see S. Lepadatu, Journal of Applied Physics 120, 163908
(2016). The x-axis ends of the magnetic wire should have electrodes so a uniform
current density is achieved – remember you can use the setdefaultelectrodes
command. You can leave the Robin heat transfer coefficient (see ambient command
output) to the default value as this is appropriate for ventilated air surrounding.

The SiO2 substrate may be added by using the addinsulator command and enabling
its heat module. You will also need to enter appropriate values for thermal
conductivity, specific heat capacity and density, and similarly for the Ni80Fe20
magnetic wire.

Note, typically the thermal cellsize can be greater than the magnetic (or electric)
cellsize, and again can be set independently in different meshes (composite media
boundary conditions do not require the discretizations to match on the two
contacting meshes, for any computational routines used in Boris; generic composite
media boundary computational routines are used which are second order accurate in
space for all meshes). For the purposes of the next exercise you can use a simple
cubic cellsize with a 10 nm side for all the thermal discretization lengths (note, if the
mesh thickness is 10 nm then the z cellsize will be adjusted so there are at least 2
computational cells along the z direction – 3D solver used).

91
Exercise 16.1

Create a Ni80Fe20 nanowire with 160 nm width, 10 nm thickness and 640 nm length,
centered on a SiO2 substrate with 800 nm width, 640 nm length and 150 nm depth.
Enable heat equation computations in both meshes and transport module in the
Ni80Fe20 nanowire. By setting appropriate insulating boundary conditions for heat
conduction, define the nanowire to be effectively infinite along the x axis, and the
substrate elongated in the x-y plane and depth. Set the ambient temperature (as well
as the base temperature – the starting temperature) to be the room temperature
value (297 K). The default thermal conductivity, specific heat capacity and mass
density values are appropriate for Ni80Fe20. For SiO2 you should edit these as K =
1.4 W/mK (thermK), C = 730 J/kgK (shc), and  = 2200 kg/m3.

a) Set a voltage step with 50 ns duration which results in a current density of

1012 A/m2 at T = 297 K. Use a temperature dependence for the electrical
conductivity  such that:

0

1  0.025T

The above formula represents the default temperature dependence set for
electrical conductivity - see paramstemp command. Obtain the average
temperature and current density as a function of time in the Ni80Fe20 mesh
both for the heating cycle (first 50 ns), as well as the next 50 ns of the cooling
cycle when the voltage is set to zero. (Note, just for this part, to speed up the
computations you may want to disable any magnetic computations in the
Ni80Fe20 nanowire by disabling the demag, exchange, and zeeman modules).

b) Set a transverse domain wall in the center of the nanowire through the
preparemovingmesh command and relax it. Redo the simulation in part a),
but this time also obtain the domain wall displacement as a function of time
when using the LLG-STT equation (use the ode command).

92
 c) Repeat part b) but this time use the LLB-STT equation with a Curie
temperature of 870 K. Note, you will need to reduce the time step significantly
for the LLB equation for numerical stability.

Figure 16.1 – a) Geometry used for Exercise 16.1, showing a magnetic wire on a
SiO2 substrate with Joule heating computations enabled – heat is generated in
the magnetic wire due to an applied charge current density. b) Average
temperature in the permalloy nanowire, also showing the current density during
and after the applied voltage pulse. c) Domain wall displacement simulated using
the LLG-STT and LLB-STT equations. For the latter a Curie temperature of 870 K
was set.

a)

b) c)

93
Tutorial 17 – Spin Transport Solver

In addition to the simple Ohm’s law used to obtain the charge current density with
the transport module, Boris also integrates a 3D spin current solver based on the
spin drift-diffusion equations – see S. Lepadatu, Scientific Reports 7, 12937 (2017).
This solver allows for a number of effects to be computed self-consistently in
arbitrary multi-layered geometries and integrated with the magnetization dynamics
solver. These include the spin Hall effect (SHE), inverse SHE, CPP-GMR, spin
diffusion and non-local spin transport effects, spin pumping, as well as bulk and
interfacial spin torques calculated from the spin accumulation and composite media
boundary conditions.

The spin transport solver computes both the charge and spin polarisation current
densities, JC and Js respectively, together with the charge potential, V, and spin
accumulation, S:

S m   SHADe e   S  P  E  P  2  E  B
2
e
J C  E   D De
B B 2e en

B     B 2
JS   PE  m  DeS   SHA B εE  B2  ei  m  im  z  E   xm   ym
e e 2e i e3n

where:

Ei  m
  im.m

B  z  xm   y m .m

Here E and B are the directions of the emergent electric field due to charge
pumping, and emergent magnetic field due to topological Hall effect respectively. Js
is a rank-2 tensor such that JSij signifies the flow of the j component of spin
polarisation in the direction i. The electric field is given by E = -V and S satisfies the
equation of motion:

94
S  S S  m m  S  m  
 .J S  De  2  2  
t    2 
 sf J  

In the above equations we have a number of material constants which can be

controlled via the params command:

 De is the electron diffusion constant

 D is the diffusion spin polarisation - this term leads to CPP-GMR
 P is the charge current spin polarisation – this terms leads to Zhang-Li spin
transfer torques, among other effects
 SHA is the spin Hall angle (unitless) – the term in the equation for Js leads to
SHE, whilst the term in the equation for JC leads to the inverse SHE; Note
there are two related parameters available in Boris: SHA and iSHA. These
represent the spin Hall angle, but may be set to different values, allowing the
SHE or inverse SHE to be turned on or off in the computations by setting one
or the other to zero.
 sf is the spin flip length
 J and  are the exchange rotation and spin dephasing lengths respectively,
describing the absorption of transverse spin components (transverse to m, the
magnetization direction) within a ferromagnetic material
 n is the carrier density (m-3)
 Charge pumping and topological Hall effect may be turned on or off by setting
the pump_eff and the_eff values to 1 or 0 (disabled by default).

Bulk spin torques are included in the computations as:

De De
TS   mS  m  m  S 
 2
J 2

This term is included in the implicit LLG (or LLB) equation as (in practice this term
results in an effective field which is added to Heff):
m m 1
 m  H eff  m   TS
t t M S

95
There is a parameter in Boris, called ts_eff (see params): This is a unitless constant
which multiplies TS and is termed the spin torque efficiency, allowing bulk spin
torques to be turned off (ts_eff = 0) or fully on (ts_eff = 1).

There are two possibilities for treating composite media boundaries. The simplest
approach is to assume continuity of a flux and potential – for the spin transport solver
these are JC and V for charge transport and Js and S for spin transport. The
continuity conditions are used when modelling interfaces between two normal metals
(N) or two ferromagnets (F); they may also be used to model interfaces between a
normal metal and ferromagnet (N/F) but in this case typically the second approach is
more appropriate, based on interface spin conductances:

   
J C .n N  J C .n F   G   G  V  G   G  VS .m

J S .n N  J S .n F    
2 B
e
 
Re G  m  m  VS   Im G  m  VS 
J S .n F 

e
 G  G V .mm  G
B  
S


 G  Vm 

In the above equations G, G are interface conductances for the majority and
minority spin carriers respectively, and G is the complex spin mixing conductance.
Also V is the potential drop across the N/F interface (V = VF – VN) and VS is the
spin chemical potential drop, where VS  De /  e /  B S . These interface conditions

describe the absorption of transverse spin components at the interface, giving rise to
interfacial spin torques:

TSint erface 
g B
ed h
    
Re G  m  m  VS   Im G  m  VS 

There is also an associated interfacial spin torque efficiency constant – tsi_eff. The
above term is also included in the magnetization dynamics equation, much in the
same way as TS is. The main difference is this torque is only included in the
computational cells at the interface, where dh is the cellsize normal to the interface –
this allows correct computation of interfacial spin torques for a ferromagnetic layer of

96
given thickness t, independent of its computational discretization, since the effect on
magnetization of the interfacial spin torque is averaged over its thickness.

Spin pumping is generated at an N/F interface as:

J Spum p 
B
2

  m
 
 m 
Re g m  t  Im g t 

Here g = (h/e2)G and the pumped spin current is used in the calculation of
composite media boundary conditions by including it on the normal metal side of the
equations. As with the spin torques, there’s an associated spin pumping efficiency
parameter – pump_eff – which allows spin pumping to be turned on or off in the
computations.

When modelling N/F interfaces you may need to have different interface
conductances on different sides of a ferromagnetic layer (e.g. a Pt/Co/Ta multilayer,
where the Pt/Co and Co/Ta interfaces may need different spin mixing
conductances). For this reason, the interface conductances (G, G, and G) are not
associated just with a ferromagnetic mesh, but also appear in the list of parameters
for normal metal meshes (conductor meshes - addconductor). When an N/F
interface is defined by the contact of two meshes, the interface conductances stored
in the upper mesh are used – e.g. if the meshes are arranged in a multilayer
structure along the z direction, the upper mesh is that with a higher z coordinate. To
turn off the interface conductance approach to modelling composite media
boundaries you need to set the G values to zero for the appropriate mesh. In this
case the computations revert to using the continuity approach described above.

To enable the spin transport solver you need to have the transport module active in
the mesh you want spin transport computations and you must also select a
magnetization dynamics equation with spin accumulation (e.g. LLG-SA, LLB-SA,
etc.) – see the ode command. Using the display command you can select to display
a number of associated quantities in the mesh viewer, including S, bulk and
interfacial spin torques, x, y, and z directions for the spin current.

97
With the spin transport solver enabled you will need to pay attention to the
convergence constant for the spin accumulation solver. Similarly to the charge
potential solver, which solves a Poisson equation to obtain V within the set
convergence constant, the spin accumulation S is obtained by solving a vector
Poisson equation – this equation is obtained from the equation of motion for S in the
“steady state”, i.e. when S/t = 0. The response time-scales of m and S are
separated typically by 3 orders of magnitude (ps vs fs time-scales respectively) thus
we only require to obtain the “steady state” values for S for a given magnetization
configuration. The vector Poisson equation also uses a convergence constant and a
timeout for the maximum number of allowed sequential iterations, and these values
may be changed by using the tsolverconfig command.

Further info:
Whilst each iteration taken for the Poisson equations for V and S is relatively cheap,
typical problems may require a large number of iterations to reach convergence,
which significantly slows down computations. This is especially true in the
initialization stage when the timeout number of iterations may be reached for the first
few iterations; after this, small steps in m should result in relatively few steps in the
solution of S (and where appropriate V). A recommended general approach is to
solve for the steady state V and S values with all spin torques turned off and for a
relaxed starting magnetization configuration. After this, save the simulation (which
also saves the computed V and S), and re-enable the spin torques as required. From
this point initialization should be quicker, with any further iterations in V and S
triggered by changes in m (changing set electrode potential values can also trigger
the Poisson solvers).
Setting the convergence factors too low may result in very slow simulations as the
solvers will require a large number of iterations. You will need to determine the best
compromise between computational speed and accuracy. The default normalised
convergence values of 10-6 for V and 10-5 for S Poisson equations are set on the
side of accuracy, having been found to give accurate results in all test cases, but you
should still verify this for your particular simulation.

98
Tutorial 18 – Spin Hall Effect

Exercise 18.1

Consider a single Pt mesh with dimensions 320 nm x 320 nm and 40 nm thickness.

Compute the spin accumulation and z-direction spin current density in response to a
set potential of 10 mV with electrodes placed at the x-axis ends. Verify that S obeys
the right-hand-rule with respect to the charge current direction.
For Pt you may use  = 7×106 S/m, sf = 1.4 nm and SHA = 0.1. You may use a cubic
cellsize with 5 nm side.
Plot the y components of the z-direction spin current density and spin accumulation
along the z-axis, through the center of the Pt slab (remember the dp_getprofile
command). Verify that the following relation holds, using the plotted value of the spin
current at the center of the Pt slab:

J sz , y  B
 SHA   /
J cx e

Figure 18.1 – Computed spin accumulation for Exercise 18.1, where the charge
current density is along the negative x direction.

99
Exercise 18.2

a) Continuing from Exercise 18.1, now add a Ni80Fe20 layer with 20 nm thickness
on top of the Pt layer. Make sure to reset to default electrodes
(setdefaultelectrodes) so a uniform charge current density is obtained in
each layer. Plot the y components of the z-direction spin current density along
the z axis for both the continuous and spin-mixing conductance interface
models for a) magnetization direction along the injected spin current, i.e.
along the y axis, and b) magnetization direction transverse to the injected spin
current, i.e. along the x axis. Explain the differences between these cases.

Note, for this exercise you will have to use a smaller cellsize along the z direction.
This is due to the large gradients involved, and is normally the case when N / F
multilayers are used. You should use a cellsize of (5 nm, 5 nm, 1 nm).

Remember you can use the computefields command for this exercise, instead of
using run, since you don’t want to relax the magnetization configuration. It helps to
monitor the transport solver number of iterations and convergence error (in the data
box display the following using the data command and righ-clicking on the respective
interactive objects: v_iter, s_iter, ts_err).

b) Does the relation in Exercise 18.1 hold at the N/F interface, and why not?
Investigate this again with a spin flip length in Pt of 8 nm, checking the relation
both at the center of the Pt layer and at the interface.

100
Figure 18.2 – Spin current density in the z direction for a Pt/Ni80Fe20 bilayer, where
the magnetization in the permalloy mesh is along the y axis (longitudinal).

Figure 18.3 – Spin current density in the z direction for a Pt/Ni80Fe20 bilayer, where
the magnetization in the permalloy mesh is along the x axis (transverse).

101
Tutorial 19 – Spin Pumping and Inverse Spin Hall Effect

In this tutorial you will set-up a ferromagnetic resonance in a magnetic dot, then
investigate the generated spin Hall voltage in a Pt underlayer. Due to the motion of
magnetic moments in the ferromagnetic layer a spin current is pumped in the Pt
underlayer, where an electrical current is generated due to the inverse SHE. This
leads to charge accumulation at opposing sides of the Pt underlayer, and thus an
electrical potential is generated.

Exercise 19.1

Setup a ferromagnetic resonance (FMR) at 20 GHz excitation frequency in a

Ni80Fe20 circle with 80 nm diameter and 10 nm thickness, with a bias field applied in
the plane of the circle along the y axis.

First, find the demagnetizing factor in the plane of the circle and set the demag_N
module with demagnetizing factors Nx = Ny = N. Remember you can calculate the
demagnetizing energy for uniform magnetization (e_demag), and the demagnetizing
factor is then related to it by:
0
 dem ag  NM S2
2

For this exercise, since you are effectively using the Stoner-Wohlfarth model you can
turn off the exchange module. Calculate the FMR bias field required for resonance at
an r.f. frequency of 20 GHz. You can use Kittel’s formula applicable for elliptical
shapes for a bias field H0 along the y direction:
0  e
f  ( H 0  ( N x  N y ) M s )( H 0  ( N z  N y ) M s )
2

Using the Hfmr stage type, apply the excitation r.f. field (together with the calculated
orthogonal bias field) in the plane of the circle for a number of cycles, and record the
average magnetization. An r.f. field amplitude of 100 A/m is normally sufficient. If the
r.f. field is applied for a sufficient number of cycles, the magnetization will achieve a

102
steady state precession at resonance. Determine the number of cycles required by
examining the output average magnetization data, then reset and save the
simulation – the next time you load the simulation the FMR precession will start
directly in the steady state.

The Hfmr stage consists of the following parameters: H0x, H0y, H0z; Hrfx, Hrfy, Hrfz; r.f.
steps; r.f. cycles.

The bias field (H0) and r.f. field amplitude (Hrf) are specified using Cartesian
coordinates. The r.f. steps is the number of discretisation steps in each r.f. cycle, and
the r.f. cycles is the number of sinusoidal oscillations the r.f. field will be applied for.
To set the required 20 GHz frequency you will need to set the correct combination of
r.f. steps and time stopping condition for each step. For example, since at 20 GHz
each period takes 50 ps, if you use 20 r.f. steps per cycle, the time stopping
condition for each step should be 2.5 ps (the default time stopping condition of 50 ps
results in a 1 GHz frequency with 20 r.f. steps per cycle, so you will need to edit this).

Exercise 19.2

Using the prepared simulation from Exercise 19.1, add a Pt underlayer with
dimensions 160 nm × 160 nm with the magnetic dot centered, and 20 nm depth.
Enable spin pumping (pump_eff = 1 in the permalloy mesh), and inverse SHE (SHA
= iSHA = 0.1) in the Pt mesh. Do not set any electrodes but make sure the transport
module is enabled in both the permalloy and Pt meshes, and the LLG-SA equation is
selected so the spin transport solver is enabled. You will need to refine the electric
cellsize in both meshes along the z direction to 1 nm. You should also relax the
transport solver convergence criteria to 10-4 for both the charge and spin solvers.

Obtain the induced spin Hall voltage at the opposing y-axis sides of the Pt mesh and
plot them as a function of time for a few FMR precessions. Note, in the output data
(data) you will have to add <V> (the average calculated voltage) for the Pt mesh two
times, editing the respective rectangles to correspond to the required two sides of
the Pt mesh.

103
Figure 19.1 – Inverse spin-Hall effect voltage in a Pt underlayer generated through
spin pumping from a ferromagnetic dot at ferromagnetic resonance.

Figure 19.2 – a) Magnetization precession at ferromagnetic resonance with a 20

GHz r.f. field, b) inverse spin-Hall effect voltage at resonance on opposing sides of
the Pt mesh – see Figure 19.1.
a) b)

104
Tutorial 20 – Ferromagnetic Resonance

In this tutorial you will learn how to simulate ferromagnetic resonance (FMR), and re-
produce input material parameters. Following this, in the next tutorial you will
investigate how the spin torques due to spin pumping and the SHE affect the
effective damping observed. An introduction to FMR simulations was given in the
preceding Tutorial, and you must complete it before proceeding. Here we will
investigate both field-swept FMR (fixed excitation frequency) and frequency-swept
FMR (fixed bias field).

Field-Swept FMR

In Boris, FMR simulations are best done using a Python script. As you will note from
the previous tutorial, applying an r.f. field excitation requires a number of cycles for
the magnetization precession to reach steady state. For a field-swept FMR
simulation, after changing the bias field you must ensure the magnetization
precession is stable before obtaining output data. The simulation procedure is as
follows:

1) Set bias field value and run the simulation for a fixed number of r.f. cycles (the
“chunk” – e.g. 20 or more), but do not save any output data.
2) After the chunk has completed, run the simulation for a single r.f. cycle and
save the output data (<M>).
3) From the saved data obtain the magnetization oscillation amplitude along the
r.f. field direction.
4) Compare the oscillation amplitude against the previous oscillation amplitude
(which is zero if this is the first chunk). If the change exceeds a set threshold
(e.g. 0.1%) then repeat from step 1), otherwise proceed.
5) Record the oscillation amplitude and bias field. Increase bias field value and
start again from step 1) until the field sweep range is completed.

105
A general-purpose FMR simulation Python script has been prepared and saved in
the examples folder for this Tutorial.

Exercise 20.1

Using a Ni80Fe20 square of 80 nm side and 10 nm thickness simulate an FMR peak

around the resonance bias field and plot the resulting magnetization oscillation
amplitude against bias field data. Set the bias field along the –y direction, i.e. at 270
azimuthal angle. Use a Python script to simulate this as described above. (You may
use Nx = Ny = 0.12, with the predicted resonance field of H0  367 kA/m; aim for at
least 50 kA/m either side of resonance).

From the simulated oscillation amplitude versus bias field data, you will need to
obtain a quantity proportional to the absorbed FMR power. The simplest way to do
this is to square the oscillation amplitude data. The FMR power absorption peak is
described by a Lorentz peak function, and you will need to fit this to your squared
amplitude data.

Boris has built-in data processing command to help with processing FMR simulation
data. You will need the following commands:

First load bias field and oscillation amplitude from the raw output data file (e.g.
named ‘fmr_fieldsweepFMR_data.txt’):

dp_load fmr_fieldsweepFMR_data 0 1 0 1

Next square the magnetization oscillation amplitude data:

dp_muldp 1 1 1

Finally fit a Lorentz peak function to the data:

dp_fitlorentz 0 1

106
The Lorentz peak function is given as:

w
f ( x)  y 0  S
4( x  x0 ) 2  w 2

In the above equation w is the full-width half-maximum (FWHM), and x0 is the peak
center. You can obtain these values from the dp_fitlorentz command, including
fitting uncertainties (Boris has a built-in generic Levenberg-Marquardt algorithm for
curve fitting).

The magnetization damping value is related to the full-width half-maximum (H) by:

 0  e H

4 f

Exercise 20.1 continued

Process the output FMR data and verify the damping obtained from the FWHM
matches the set damping value (damping = 0.02).

Figure 20.1 – Simulated FMR peak with Lorentz peak function fit for Exercise 20.1.

107
Frequency-Swept FMR

Frequency-swept FMR simulations are also possible and are typically much more
efficient to compute than field-swept FMR. This type of simulation is closely related
to the method used to investigate spin-wave dispersion. Here we apply a sinc pulse
in the time domain and capture the magnetisation response, which we then
transform to the frequency domain using a Fourier transform. The reason for using a
sinc pulse is its Fourier transform is a symmetric hat function with a defined
frequency cut-off. Thus in order to capture the required resonance (and also any
required higher resonance modes) we simply need to set the cut-off to a large
enough value.

The sinc pulse excitation is given as:

H (t )  H e sin 2f c t  t0  / 2f c t  t0 

Here fc (Hz) is the cut-off frequency, He is the excitation amplitude, with H(t) taking
on the value He at t = t0, the sinc pulse centre. We also need a fixed bias field, H0,
which must be orthogonal to the excitation field, and in general we must capture the
average magnetisation (for now we’ll assume the magnetisation is uniform) and use
the magnetisation component along the excitation field. We need to capture the
magnetisation at a time step set by the Nyquist criterion:

1
tS  (s)
2 fC

Capturing magnetisation data at other time steps is sub-optimal and will result in lost
information if larger than the above formula, or increased noise in the Fourier
transform spectrum if smaller than the above formula. The sinc pulse must be
simulated for a total time of 2t0: simulating for longer or shorter than this time will
result in increased noise in the Fourier transform spectrum. If you want to increase
the spectrum resolution (number of points) then you must instead increase the t 0
value, but still simulate for a total time of 2t0, saving data at a time step of ts.

108
We can set this type of excitation using a text equation stage, namely Hequation. To
set the above equation, edit the stage value for the added Hequation stage to:
He*sinc(2*PI*fc*(t-t0)), H0, 0. This will set the excitation field along the x axis, and
bias field along the y axis, with zero z axis field. A dedicated chapter in the manual is
given for details on using text equations. In the above equation t is a reserved
parameter, namely the stage time, and the equation provided is evaluated internally
every iteration. The remaining parameters, He, fc, t0, H0 must be defined in order for
the equation evaluation to function. Equation constants may be given as:

equationconstants name value,

e.g. equationconstants fc 200e9.

Other related commands are delequationconstant, and clearequationconstants.

For frequency-swept FMR after taking the Fourier transform, similarly to field-swept
FMR, the output data must be squared. The resulting FMR peak in the frequency
domain is also described by a Lorentz peak, where the FWHM, Δf, is related to the
resonance frequency f0, and Gilbert damping as:

f

2 f0

By capturing a set of frequency-domain FMR peaks as a function of bias field H0 the

following Kittel formula can be verified:

0  e
f  H  N  N z  k M S H 0  N x  N z  k M S 
2
0 y

Here Nx, Ny, and Nz are demagnetizing factors such that the bias field is applied
along the z direction. The above formula also includes uniaxial anisotropy, with easy
axis along the z direction, where:
2 K1
k
 0 M S2

109
For the following exercise you’ll simulate a thin film Co material interfaced with Pt
(this is stored in the materials database and can be loaded as setmaterial Co/Pt –
more on the materials database in a dedicated chapter. You will also need to
simulate a thin film, thus must set periodic boundary conditions in the xy plane as:
pbc x 10, pbc y 10 – more on periodic boundary conditions in a dedicated tutorial.

Exercise 20.2 (Advanced)

Simulate the frequency-swept FMR response of a thin-film Co/Pt material with 2 nm

thickness using the method described above.

You can use a cut-off frequency of 200 GHz, with excitation field amplitude of 1000
A/m, and vary the bias field between 100 kA/m and 1 MA/m. Simulate both out-of-
plane FMR (anisotropy easy axis and bias field out of plane), and in-plane FMR
(anisotropy easy axis and bias field in the plane, e.g. both along the y direction).

In both cases use the Kittel relation and damping formula to verify the input
simulation parameters using fitting procedures (damping and magneto-crystalline
anisotropy).

110
Tutorial 21 – Ferromagnetic Resonance with Spin Torques

Continuing from the previous tutorial, you will now investigate the effect of spin
torques due to the spin-Hall effect on the magnetization damping using a Pt/Ni80Fe20
bilayer. The spin current generated in the Pt underlayer is absorbed by the Ni 80Fe20
layer, resulting in a combination of damping-like and field-like torques. Depending on
the current direction a decrease or increase of the effective damping is obtained.
First the full spin transport solver is used, and following this a simpler method using
an analytical form for the spin-orbit torques is introduced. Using the full spin transport
solver we can also consider the effect of spin pumping on the effective damping, and
this is investigated at the end of this tutorial.

The interfacial spin orbit torque added to the implicit LLG equation, as explained in
Tutorial 17, is given by:

TSint erface 
g B
ed h
    
Re G  m  m  VS   Im G  m  VS 

For an N/F interface in the x-y plane with uniform current densities we can obtain an
analytical expression for this interfacial torque as (see S. Lepadatu, Scientific
Reports 7, 12937 (2017)):

B Jc
TSint erface   SHA, eff m  m  p  rG m  p
e dh

Here p = z × eJc, where eJc is the charge current direction, and:


 SHAeff   SHA 1 
1 
2
 2

 Re G~  Im G~  N  Re G~
 cosh(d N / sfN ) 
   
~ 2
 N   Re G  Im G 
~2 ,

~ ~ ~
N  Im{G}  2 Re{G} Im{G}
rG  ~ ~ ~
N  Re{G}  Re{G}2  Im{G}2

111
In the above, N   tanhd N /  sfN /  sfN , F  tanhd F / Fsf / Fsf , and G  2G  /  N . Thus
~

the interfacial torque has a damping-like and a field-like component. In the limit of
abrupt interface (φ  0 or equivalently Re{G}  ) the field-like component tends
to zero and we obtain the following expression for the torque:

B Jc
TSOT   SHA, eff m  m  p 
e dh
and
 1 
 SHA, eff   SHA1  
 cosh(d N / sfN ) 


This approximation can also be used when the damping-like torque is much larger
than the field-like torque. This expression is commonly used in the literature to model
the spin-orbit torque resulting from the spin-Hall effect. Note however the spin-Hall
angle in this expression is not the bulk (or intrinsic) spin Hall angle, but an effective
spin Hall angle, scaled by transport parameters; if further the N layer thickness is
many times larger than its spin flip length, we can use the approximation
 SHA, eff   SHA . In many cases this may not be true, and moreover the abrupt interface
approximation may not be good either, thus to model the effect of the damping-like
torque with the analytical form of the spin-orbit torque, in the expression for TSOT you
should use the full expression for the effective spin-Hall angle given above.

In Boris you can include this analytical spin-orbit torque using the SOTField module.
Enabling this module in a ferromagnetic mesh introduces an additional effective field
into the LLG equation which results in the TSOT torque given above (as it appears in
the implicit LLG equation). To use it you still need to have the transport module
enabled in order to calculate the charge current density, but instead of selecting
LLG-SA (enabling the full spin transport solver) you should select just the LLG
equation (ode command). In the material parameters for the ferromagnetic mesh
(params command) you need to enter the correct effective spin-Hall angle (SHA) to
use with the SOTField module.

112
Exercise 21.1

Using the Ni80Fe20 layer from the previous tutorial, now add a Pt underlayer with the
same dimensions (80 nm × 80 nm × 10 nm), using the Pt parameters from Tutorial
18. Set default electrodes (resulting in current flow along the x direction), enabling
the spin transport solver both in the Ni80Fe20 and Pt meshes (add transport modules
and set the ode solver to LLG-SA). Make sure to disable spin pumping (pump_eff =
0) and the inverse SHE (iSHA = 0). You should also disable bulk spin torques (ts_eff
= 0), only leaving interfacial spin torques enabled (tsi_eff = 1). As before you may
need to decrease the z-direction electrical cellsize to ensure accuracy (and
numerical convergence!).

Obtain FMR peaks for charge current densities in the Pt layer of Jc = 1012 A/m2.
How does the damping change with current density direction?

In this case, even though you are using the Stoner-Wohlfarth model (demag_N
module), you should still enabled the exchange module. The reason for this, the spin
torques may not be perfectly uniform (e.g. if the permalloy and Pt layers have the
same width, the spin torques will not be uniform since the spin accumulation has
gradients at the sample edges), thus you do need to take the exchange interaction
into consideration.

The change in damping due to a damping-like spin-orbit torque may be roughly

approximated by:

B Jc
 SHE   SHA,eff
e 2fM S d F

Verify the change in damping obtained from simulations with the above formula.

113
Exercise 21.2

Repeat Exercise 21.1 but this time without the spin transport solver, only using the
analytical form for the spin-orbit torque (SOTField module). You should delete the Pt
mesh and reset the electrodes and potential to give you the correct current density.
Calculate an appropriate effective spin-Hall angle to use. Compare the results with
the previous exercise.

Exercise 21.3

Repeat Exercise 21.1, using the full spin-transport solver, but now enable spin
pumping (set pump_eff = 1). What is the increase in damping?

Figure 21.1 – FMR simulations with spin orbit torques for both the full spin transport
solver (ST Solver) and effective field obtained from the analytical spin-orbit torque
(SOTField).

114
Tutorial 22 – CPP-GMR

The spin transport solver is also able to reproduce the spin torques in a current-
perpendicular to plane (CPP) giant magneto-resistance (GMR) spin valve, in addition
to its magneto-resistance. Here we will investigate the current-induced switching in a
simple generic spin valve between the parallel and anti-parallel states, see Figure
22.1, and plot the resistance during these switching events.

A spin valve, in its simplest form, consists of a fixed magnetic layer, a free magnetic
layer which can be switched between an anti-parallel and parallel orientation with
respect to the fixed layer, and a thin metallic spacer layer. The spacer layer
thickness can be adjusted to give either a ferromagnetic or anti-ferromagnetic
surface exchange coupling between the two magnetic layers. In the following
simulation we will also add two metallic contacts, top and bottom.

Figure 22.1 – CPP-GMR spin valve showing the spin accumulation in the spacer
layer, top, and bottom contacts, and the magnetization in the elliptically shaped fixed
and free layers for a) anti-parallel state, and b) parallel state.

115
Exercise 22.1

Setup a generic spin valve structure (i.e. just use the default mesh parameters
unless indicated otherwise) similar to that shown in Figure 22.1. This consists of:

 Bottom and top contacts (addconductor) with dimensions 160 nm × 80 nm ×

20 nm. Disable spin-Hall effects in both (SHA = iSHA = 0).
 Spacer layer with dimensions 160 nm × 80 nm × 2 nm and set it to an elliptical
shape (drag a .png file with a circle shape to the mesh viewer when the
spacer layer mesh is in focus). Disable spin-Hall effects.
 Fixed layer (addmesh) with dimensions 160 nm × 80 nm × 10 nm and
elliptical shape. Disable spin torques and spin pumping in this mesh (ts_eff =
tsi_eff = ts_pump = 0). You should also disable magnetization dynamics in
this mesh so the magnetization is fixed. You can do this by setting the relative
gyromagnetic factor to zero (grel = 0 in params).
 Free layer with dimensions 160 nm × 80 nm × 5 nm and elliptical shape.
Disable spin pumping and bulk spin torques only, in this mesh (i.e. keep
tsi_eff = 1).

You will need to add the following modules:

 super-mesh multi-layered demagnetization (sdemag).

 surfexchange modules in both magnetic meshes. Edit the J1 (bilinear surface
exchange energy density) value in the free layer to give you a weak
ferromagnetic coupling; set J1 = 0.1 mJ/m2.
 transport modules in all meshes. Set the electrical cellsize to 5 nm × 5 nm × 1
nm everywhere except in the spacer layer where you should set it to 5 nm × 5
nm × 0.5 nm.

For the ode solver you should set the LLG-SA equation (thus enabling the spin-
transport solver) with RKF45 evaluation. For output data you should have time, R
(resistance), and <M> (average magnetization) in the free layer. Set electrodes top
and bottom (addelectrode), designating the bottom electrode to be the ground
electrode (electrodes). You need to simulate switching starting from the anti-parallel

116
state (see Figure 22.1) using a +15 mV pulse for 3 ns, then back to this state with a
further -15 mV pulse for 3 ns. Save data every 10 ps for both stages. Before starting
the simulation you should relax the starting state as follows:

1) Set all spin torques to zero and insert a Relax stage with nostop condition at
the start.
2) First relax the magnetization in the anti-parallel state without the spin-
transport solver (set ode to LLG).
3) Next enable the spin-transport solver and relax it (run it until the solver no
longer iterates, monitoring v_iter and s_iter data).
4) Re-enable the appropriate spin torques (tsi_eff = 1 in the free layer only),
reset, delete the Relax stage, then save the simulation (savesim).

Explain the resistance change observed by comparing it with the magnetization in

the free layer as a function of time – see Figure 22.2 for expected results.

Figure 22.2 – Change in resistance for the CPP-GMR spin valve of Exercise 22.1,
together with the magnetization along the longitudinal direction in the free layer.

117
Tutorial 23 – Skyrmion Movement with Spin Currents

Skyrmions may be displaced efficiently using charge and spin currents. To study
their movement a skyrmion tracking window can be used in Boris. There are two
methods available for tracking a skyrmion, discussed below, both available as data
outputs: skyshift and skypos (use data command).

The skyshift entry needs a rectangle defined, which should be set around the initial
position of a skyrmion, making sure to fully contain it, but don’t leave excessive
space around it; the thickness of this rectangle should be set to the thickness of the
ferromagnetic mesh containing the skyrmion. During a simulation a x-y shift is
recorded and saved in the output data file. This shift is determined by comparing the
average magnetization magnitude in the 4 quadrants of the skyrmion tracking
window – e.g. if the skyrmion shifts to the right, the average magnetization
magnitude in the 2 right-hand-side quadrants will decrease compared to the left, thus
a right single-cell shift is recorded. Multiple skyshift entries can be defined, with
different rectangles, to track multiple skyrmions. Note, the skyshift entry only works
with data file output, and not in the data box or with the showdata command.

The raw output skyshift data will contain staircase steps due to mesh discretisation.
It is possible to obtain a more natural skyrmion movement path by assuming linear
displacement in between the staircase steps – this is illustrated in Figure 23.1. Here
skyrmion displacement was simulated for 3 ns and the individual x and y skyshift raw
data are shown in Figure 23.1(a). To remove the stair steps and replace them using
linear interpolation you can use the dp_replacerepeats command on both the x and
y skyshift data columns. The x, y data can then be plotted directly in Cartesian
coordinates. You will notice this path doesn’t start from (0, 0). To display the
skyrmion displacement path relative to its starting position you should remove this
offset using the dp_removeoffset command on both the x and y data. Finally, if you
want to plot this path using polar coordinates you can use the dp_cartesiantopolar
command, included in Boris for convenience. Note, especially when converting to
polar coordinates you should check the processed data correctly represents the raw
data. Problems may occur due to blips in the raw data, especially if the tracking

118
window was not defined well or the starting state is not sufficiently relaxed, thus the
results from this procedure must be carefully compared with the raw data.

Figure 23.1 – Skyrmion movement raw data processing, showing (a) individual x
and y displacements, (b) Cartesian coordinates path, and (c) polar coordinates path.

The second method uses the skypos data output. Again this needs an initial
rectangle defined around the skyrmion as for the skyshift data output. Skypos uses a
far more computationally expensive algorithm to track the skyrmion, but is able to
obtain the exact skyrmion center position, as well as skyrmion diameters along the x
and y axes, without being affected by staircase discretisation artifacts. It is also able
to adjust the tracking window size if the skyrmion diameter changes significantly. The

119
algorithm works by fitting the skyrmion using an analytical function (the same one
used for the dp_fitskyrmion command) in three steps: 1) Skyrmion is fitted along
the x axis through the center of the skyrmion tracking window in order to find the
center position. The tracking window x position is adjusted to center it. 2) Skyrmion is
fitted along the y axis through the center of the skyrmion tracking window in order to
find the center position. The tracking window y position is adjusted to center it. The y
diameter is also recorded at this step. 3) The skyrmion is again fitted along the x axis
to obtain its x diameter.

For the following exercises you should first use the skyshift method of tracking the
skyrmion, the repeat them using the skypos method. Note, you should not use both
methods simultaneously on the same skyrmion.

Exercise 23.1

a) Setup a Pt/Co bilayer with a skyrmion relaxed at the center of the Co layer
under a 15 kA/m out-of-plane magnetic field as shown in Figure 23.2. The Pt
layer should be a 320 nm × 320 nm × 3 nm rectangle, whilst the Co layer
should be a 320 nm diameter disk with a 1 nm thickness. Relax this
magnetization configuration.

For Pt you should use  = 7×106 S/m, sf = 1.4 nm and SHA = 0.19. Use a
discretisation cellsize of (4 nm, 4 nm, 0.5 nm). Set iSHA to zero.

For Co you should use  = 5×106 S/m, sf = 38 nm, J = 2 nm,  = 4 nm, Gmix
= 1.5 PS/m2, grel = 1.3,  = 0.03, MS = 600 kA/m, A = 10 pJ/m, D = -1.5 mJ/m2,
K1 = 380 kJ/m3 with uniaxial anisotropy perpendicular to the plane. You
should also enable the interfacial DM exchange module. Use a discretisation
cellsize of (4 nm, 4 nm, 1 nm) for magnetic computations and (4 nm, 4 nm,
0.25 nm) for spin transport computations. In the Co mesh only enable the
interfacial spin torques, not the bulk spin torques or spin pumping.

120
Figure 23.2 – Skyrmion in a Co disk on a Pt underlayer.

b) Enable the spin transport solver in both meshes and set electrodes at the x-
axis ends of the Pt mesh only. Set a -20 mV potential for 3 ns and save the
time and skyshift (or skypos) data every 10 ps. (for skyshift and skypos define
a rectangle around the initial position of the skyrmion). Simulate the skyrmion
movement path with SHE enabled (SHA = 0.19 in the Pt mesh), as well as
without SHE (SHA = 0 in the Pt mesh), and plot them in polar coordinates.

c) Simulate the skyrmion movement path without the spin transport solver but
with the SOTfield module enabled. Still keep the transport module enabled to
calculate the charge current density. For the SOTfield module set a suitable
effective spin Hall angle in the Co mesh (SHA). You can use the effective spin
Hall angle formula from Tutorial 21, but note this is only strictly applicable for

121
 uniform magnetization and spin currents. You can use this as a starting point,
but will need to adjust the effective spin Hall angle.

Plot the skyrmion path in polar coordinates and compare it with the path
obtained as the difference between the SHE and no SHE simulations above –
see Figure 23.3 for expected results.

Figure 23.3 – Skyrmion movement paths obtained in Exercise 23.1.

122
Tutorial 24 – Roughness and Staircase Corrections

Staircase Corrections

With finite difference discretisation, errors can arise due to a staircase effect when
discretising curved boundaries. In micromagnetics the largest errors arise in the
demagnetizing field and may be reduced by decreasing the discretisation cellsize.
This method is inefficient however since for most problems the results converge
when the discretisation cellsize is close to the exchange length of the material – thus
to further reduce this everywhere just to improve the discretisation accuracy at a
boundary is very inefficient. Note, for materials where the demagnetizing energy
dominates the exchange length may be defined as:

2A
lex 
 0 M S2

For systems where the anisotropy energy dominates (Ku > µ0MS2/2), the exchange
length may be defined as:

A
lex 
Ku

Instead of refining this, a good approximation may be achieved by computing a

correction field using a finely discretised demagnetization kernel, but applying it at
run-time to the coarsely discretised mesh, as described in S. Lepadatu, Journal of
Applied Physics 118, 243908 (2015). This correction field is typically similar to an
uniaxial anisotropy field when averaged.

To enable staircase corrections in a particular magnetic mesh you must enable its
Roughness module. When applying a mask shape to the mesh, staircase corrections
will now automatically be taken into account. You must enable the Roughness
module and reset the mesh shape before applying the mask to correctly enable

123
staircase corrections. You must also set the required refinement using the
refineroughness command:

refineroughness mx my mz

When calculating the correction field factors, the shape is first discretised on a fine
mesh as set by the refineroughness parameters. Thus if the coarse mesh has
cellsize (hx, hy, hz), the fine mesh used for correction field initialization has cellsize
(hx / mx, hy / my, hz / mz). Typically the improvement in accuracy is small above m >
10, so the refinement set should not be excessive. Since a demagnetizing kernel
must be computed for the fine mesh, the initialisation time may become very long,
and the available memory may be exceeded if the m factors are set too large. You
must also set them before applying the mask shape.

With the Roughness module enabled, setting a mask shape may result in a slightly
different shape than without. This is because a fine shape is internally obtained first,
then the coarse mesh shape is calculated to be the smallest shape which includes
the fine shape on the coarse mesh – this is a requirement of the corrections
calculation method. To clear the staircase corrections, effectively setting the fine
mesh shape to the coarse mesh shape you can use:

clearroughness

There’s an energy density term associated with the correction fields, and this is
available as a data parameter: e_rough. The demagnetizing energy density,
e_demag, still corresponds to the coarse mesh shape; the sum of e_rough and
e_demag is the approximated demagnetizing energy for the fine mesh shape.

124
Exercise 24.1

Set a 80 nm diameter Ni80Fe20 disk with 10 nm thickness. Calculate the

demagnetizing energy density as a function of in-plane uniform magnetization
orientation from 0 through 360 by saturating in a strong magnetic field (106 A/m).
Repeat this computation but now set the shape with the Roughness module enabled
and a refinement of (10, 10, 1) – refineroughness. Compare the demagnetizing
energies for the coarse mesh and the approximated demagnetizing energy for the
fine mesh (e_rough + e_demag).

For the above exercise, in theory the demagnetizing energy should be constant for a
circle as the field rotates. In practice, due to discretisation errors a shape anisotropy
effect is observed (the magnetization is not fully saturated even at 106 A/m, so some
non-uniformity persists). With staircase corrections enabled this anisotropy should be
significantly reduced, thus closer to the ideal uniform demagnetizing energy – see
Figure 24.1. The refinement can be increased but further improvement is small.

Figure 24.1 – Demagnetizing energy computed for a circle with and without
staircase corrections.

125
Edge and Surface Roughness

The same model used to reduce staircase corrections may be applied to compute
the effect of topological roughness with variations below the exchange length of the
material. As before, coefficients for a roughness field are computed at initialisation
depending on the shape of a finely discretised mesh (the mesh with topological
roughness applied), and that of the coarse mesh (the actual mesh used in
computations but without roughness).

A roughness profile may be applied using a built-in algorithm, or alternatively a mask

may be used. See Figures 24.2 – 4 for examples of real surface scans, processed
into a grayscale image suitable for use as masks. These may be found in the
Examples folder for this tutorial.

Figure 24.2 – Granular surface roughness profile

126
Figure 24.3 – Maze-like surface roughness profile.

Figure 24.4 – Elongated defects, or stripes, surface roughness profile.

To apply a roughness profile using a mask, instead of simply dragging the file to the
mesh viewer (as you would do when applying a shape), you should also specify the
depth to which you want to apply the profile. For example, with the mask shown in
Figure 24.4, to apply it to a 4.5 nm depth (the coarse discretisation cellsize is 5 nm
so this keeps the coarse mesh shape intact) you need to use:

loadmaskfile 4.5nm (directory\)Stripes

This will apply a surface roughness profile on the top face up to 4.5 nm depth – black
results in 0 depth cut, whilst white results in full depth cut (i.e. 4.5 nm); values on the

127
greyscale in between are correlated linearly with the depth cut. For the actual
surface roughness profile obtained see Figure 24.5.

In this example the starting mesh has dimensions of 320 nm × 320 nm × 10 nm, and
the roughness refinement was set to (4, 4, 10) – refineroughness. To view the set
roughness, under display select the Roughness option for the respective mesh. To
apply the surface roughness to the bottom face, negative values need to be set for
the depth value – see help for loadmaskfile command.

Figure 24.5 – Applied surface roughness using the mask in Figure 24.4.

You can also apply edge and surface roughness using a built-in console command.
Currently two methods are available: roughenmesh, surfroughenjagged. The
roughenmesh command applies a completely random roughness profile to one of
the 6 faces as indicated – see help for this command. The surfroughenjagged
command applies a jagged profile to either the top, bottom, or both, faces – see help
for this command.

128
Exercise 24.2

Set a 160 nm × 160 nm × 10 nm permalloy rectangle and apply the Stripes

roughness profile from Figure 24.4 (use file in Examples folder) to a depth of 4.5 nm.
Use a roughness refinement of (4, 4, 10). Simulate the hysteresis loops along the x
and y directions and compare them. (You should apply the field at a slight angle to
the x and y directions to avoid artifacts associated with a finite geometry – in
particular for the easy axis you want to avoid the “U” shape configuration at zero field
which can happen if the field is perfectly along the x axis.)

Since permalloy does not have a magneto-crystalline anisotropy, without roughness

it is expected the two hysteresis loops will be identical. With roughness applied an
effective anisotropy is observed, due to the orientation of the surface roughness
stripes as seen in Figure 24.5.

Figure 24.6 – Hysteresis loops for Exercise 24.2, showing a roughness-induced

anisotropy effect.

129
Tutorial 25 – Defects and Impurities

Material parameters in Boris may also be assigned a spatial variation, in addition to a

temperature dependence. This spatial variation will be taken into account in all
routines where the material parameters appear, allowing inclusion of material defects
and impurities in simulations as appropriate.

To see the currently set parameters spatial variation use the command:

paramsvar

You can use a pre-defined method of generating defects by following the instructions
displayed after using the paramsvar command. Currently these include: random,
jagged, defects, faults. To see the spatial variation generated, under display select
the ParamVar option, making sure to select the required parameter under the
paramvar list. The generated spatial variation is stored as an array of coefficients,
multiplying the base parameter value. For examples of these profiles see Figures
25.1 – 4.

Figure 25.1 – Random parameter variation between 0.9 and 1.1 with generator seed
1.

130
Figure 25.2 – Jagged parameter variation between 0.9 and 1.1 with 30 nm average
spacing and generator seed 1.

Figure 25.3 – Defects parameter variation between 0.9 and 1.1 with diameters in the
range 20 nm to 50 nm, and 40 nm average spacing with generator seed 1.

131
Figure 25.4 – Faults parameter variation between 0.9 and 1.1 with 20 nm to 50 nm
fault length, -30 to 30 fault orientation and 50 nm average spacing with generator
seed 1.

Exercise 25.1

Generate MS defects in a 640 nm × 640 nm × 10 nm mesh as shown in Figures 25.1

– 4.

You can also set a custom parameter variation using an image as a mask file,
although this is now a legacy option and documented in a previous version (v2.4).
Instead if you want to set an arbitrary parameter variation you should use the ovf2
file option, which can be programmatically generated – a routine is included in the
NetSocks module, allowing for easy generation of parameter spatial variation. This is
covered in a separate chapter in the manual on working with ovf2 files. You can also
set parameter variation using a text equation, which simultaneously allows for both
spatial and temporal dependence. This is covered in a separate chapter in the
manual on working with text equations.

132
Tutorial 26 – Polycrystalline and Granular Films

Boris includes a Voronoi tessellation generator, both 2D and 3D, which can be used
to generate polycrystalline and granular films.

Figure 26.1 – Polycrystalline film showing K1 parameter variation generated using

vor2D generator between 0.9 and 1.1 with 40 nm spacing and generator seed 1.

Figure 26.2 – Polycrystalline film showing easy axis (ea1) parameter variation
generated using vorrot2D generator with polar angle range 70 to 110, azimuthal
angle range -90 to 90, with 40 nm spacing and generator seed 1.

133
Polycrystalline films may be simulated by generating parameter variations using one
of the Voronoi tessellation generators under the paramsvar command. These
include:

vor2d min, max; spacing; seed – Used for 2d crystallites in the xy plane.
vor3d min, max; spacing; seed – Used for 3d crystallites.

vorbnd2d min, max; spacing; seed – Used for 2d crystallites in the xy plane, but
parameter variation generated randomly only at Voronoi cell boundaries.
vorbnd3d min, max; spacing; seed – Used for 3d crystallites, but parameter
variation generated randomly only at Voronoi cell boundaries.

vorrot2d min_polar, max_polar; min_azimuthal, max_azimuthal, spacing; seed –

Used for 2d crystallites in the xy plane, specifically magneto-crystalline anisotropy
easy axes.
vorrot3d min_polar, max_polar; min_azimuthal, max_azimuthal, spacing; seed –
Used for 3d crystallites, specifically magneto-crystalline anisotropy easy axes.

Both 2D and 3D crystallites may be generated using the generators listed above. For
an example of a 2D polycrystalline film with K1 (magneto-crystalline anisotropy)
variation see Figure 26.1. In order to generate crystallites with a varying magneto-
crystalline anisotropy easy axis orientation you can use either the vorrot2d or
vorrot3d generator – for example see Figure 26.2 for the ea1 parameter having the
same polycrystalline structure as in Figure 26.1. Figure 26.2 shows the rotation to be
applied, as a vector quantity. For an easy axis base value set along the x-axis this
coincides with the resulting easy axis orientation.

You can also generate a parameter variation at the Voronoi cell boundaries rather
than in the cells themselves. This can be done using the vorbnd2d and vorbnd3d
generators. This could be useful for example to modify the electrical conductivity at
the grain boundaries only.

In order to generate a granular film with non-magnetic phase separation you can use
one of the following commands:
134
generate2dgrains spacing (seed)
generate3dgrains spacing (seed)

These commands generate granular films directly on the magnetization mesh – see
for example Figure 26.3. You can also combine this with parameter variations
generated using the same generator seed and sizes (e.g. grains with varying MS
values).

When generating grains for the magnetic mesh you can choose to generate grains
for the electrical conductivity mesh also. To carry the grain structure over you should
generate the grains first without the Transport module enabled. After enabling the
Transport module the granular structure is also applied to the electrical conductivity
mesh. This could be useful for example in a multi-layered structure. If you apply the
grain structure with the Transport module already enabled, the grains are not
generated for the electrical conductivity also. You could combine this with a Voronoi
generator for the elC material parameter however (e.g. vorbnd2d or vorbnd3d, or
even vor2d or vor3d) as mentioned above.

Figure 26.3 – Granular film (80 nm thick) with non-magnetic phase separation,
generated using generate3dgrains command with 50 nm spacing and generator
seed 1. The image shows a magnetization configuration at zero field.

135
Tutorial 27 – Periodic Boundary Conditions

Periodic boundary conditions (PBC) may be applied when computing demagnetizing

fields. Enabling this setting also affects exchange coupling (e.g. Exchange,
DMExchange, iDMExchange), resulting in a wrap-around effect. PBCs are useful to
simulate periodic arrays, or to approximate effectively infinite thin films or tracks
using only a finite simulation window. PBCs are applicable to both single mesh
demagnetization (Demag), as well as supermesh demagnetization (SDemag),
including multi-layered demagnetization. Moreover, when enabling the Roughness
module, PBCs are also taken into account when calculating an effective roughness
field.

To enable PBCs use the command:

pbc

The configuration for all meshes, or the supermesh if applicable, will be shown. You
can enable PBCs in any direction by setting a number of images using the interactive
console objects (10 images are set by default when enabled, which can be edited if
required; the default setting is for no PBCs). If the SDemag module is enabled you
can edit the PBC settings on the supermesh, otherwise you need to edit them for the
individual meshes.

When using PBCs with multi-layered convolution you need to ensure the problem is
physically meaningful. For example the recommended approach is to use a z
direction stacking of layers, then either x or y PBCs are fine, but z PBCs will not lead
to physically meaningful results in this case.

136
Exercise 27.1

Repeat Exercise 24.2, but this time set default PBCs along both x and y. Compare
the hysteresis loops.

When using PBCs with the Roughness module enabled, you should be careful about
setting a combination of a large number of PBC images and fine roughness
refinement (refineroughness). This can result in excessive initialization time due to
the large demag kernel calculated by the Roughness module.

Figure 27.1 – Hysteresis loops obtained for Exercise 27.1. The small peaks in the
hard axis loop are artifacts resulting from the finite simulation mesh size, and could
be decreased by increasing the simulation mesh rectangle.

137
Tutorial 28 – Ultrafast Demagnetisation

Ultrafast demagnetisation processes may be studied in Boris using the LLB equation
(or sLLB) coupled to a two-temperature model. The default heat equation doesn’t
differentiate between lattice and electron temperature, i.e. it is a 1-temperature
model. With the two-temperature model, two coupled equations for the electron and
lattice temperatures are given as:

Te (r, t )
Ce ρ  .KTe (r, t )  Ge (Te  Tl )  S
t
T (r, t )
Cl ρ l  Ge (Te  Tl )
t

Here Ce and Cl are the electron and lattice specific heat capacities,  is the mass
density, K is the thermal conductivity, and Ge is the electron-lattice coupling
constant, typically of the order 1018 W/m3K.

To change the temperature model use the following command (or use the interactive
console output from tmodel command):

tmodel num_temperatures (meshname)

e.g. tmodel 2 sets the two-temperature model for the currently focused mesh (nnot
applicable to insulator meshes). The material parameters in the above equations are
available as usual under the params command console output (also see list of
parameters under the Material Parameters section in this manual).

In the above equation S is a heat source; a heat source may be specified in

simulation using one of the following stage types: Q, Q_seq, Qequation, Qfile. These
stages specify a heat source (W/m3) in the heat equation: Q sets a constant value,
Q_seq sets a sequence of values (similar to Hxyz_seq for external fields), Qequation
sets a heat source using a text equation, thus allowing both spatial and temperature
dependence, and finally Qfile sets a spatially uniform heat source, but with arbitrary
time dependence as specified using a text file (see description of stage in console

138
help). Moreover a spatial variation may be assigned to the parameter Q under the
paramsvar command.

A typical heat source from a focused laser pulse is given as:

  | r  r0 |    (t  t 0 ) 2 
S  P0 exp  2  exp  2  (W / m 3 )

 d / 4 ln( 2)   R t / 4 ln( 2 ) 

Here d and tR are full-width at half-maximum (FWHM) values (pulse diameter and
duration respectively) and P0 is the maximum power density. This heat pulse may be
simulated using a Qequation stage set to (pulse centre coincides with simulation
mesh centre):

Q0 * exp(-sqrt((x/Lx - 0.5)^2 + (y/Ly - 0.5)^2) / ((d0/Lx)^2/(4*ln(2)))) * exp(-(t-

2*tau)^2/(tau^2/(4*ln(2))))

For this equation we need to define the user constants (equationconstants): i) Q0,
ii) d0, iii) tau.

Exercise 28.1

In this exercise you will simulate the effect of a single laser pulse on a Co/Pt/SiO2
trilayer, similar to that used in S. Lepadatu (2020) arXiv:2005.13238, when the
electron temperature rises above the Curie temperature, and observe creation of
Néel skyrmions, plotting the variation in topological charge magnitude as a function
of time. The structure to simulate is shown in Figure 28.1. The topological charge is
computed as:

1  m m 
Q  m. 
4 A  x y 
dxdy

The topological charge takes on unit values for skyrmions, and may be computed in
Boris using the dp_topocharge command.

139
Figure 28.1 – Trilayer structure similar to that used for Exercise 28.1, consisting of
Co (2 nm) / Pt (8 nm) / SiO2 (40 nm), showing (a) temperature during a Gaussian
profile laser pulse, and (b) typical ultrafast laser pulse and temperature time
dependence in the three layers: Co layer maximum electron and lattice
temperatures, Pt layer average electron temperature and SiO2 average temperature.
Reproduced from S. Lepadatu (2020) arXiv:2005.13238.

See the ufsky_creation.py script in the Examples/Tutorial 0 folder. This script sets up
the simulation for this exercise from scratch. Study the script to understand what all
the commands are used for (if needed look up the command help in the console or
manual). Run this script several times and build a probability distribution of number
of skyrmions created in each run. Is the computed probability distribution described
by a Poisson counting distribution? What is the mean number of skyrmions created?

140
Hint: to obtain a reasonable probability distribution you will need to run the script at
least 50 times – this skyrmion counting process can be automated, and will require
up to 1h or 2h of simulation time depending on your workstation.

Further hint: the computed Q value with dp_topocharge will not be an integer for two
reasons: i) stochasticity, ii) cellsize could be too large. However, rounding the Q
value will result in a correct integer Q value. The other possibility is to turn off the
stochasticity at the end of the simulation and allow the magnetisation to quickly relax
before obtaining the Q value with dp_topocharge. With an in-plane cellsize of 1 nm
this will result in a value very close to an integer (e.g. -3.98 for 4 skyrmions etc.).

Figure 28.2 – Topological charge magnitude as a function of time, computed for a

run of Exercise 28.1. The resulting skyrmion state after 800 ps is shown in the inset,
showing 6 skyrmions created (Q  -6).

141
Tutorial 29 – Magneto-Optical Effect

With circularly polarised pulses the polarisation can be clockwise or anti-clockwise.

Due to the circular polarisation of the laser pulse a strong perpendicular magneto-
optical field is present, given by H MO    H MO
0
f MO r, t zˆ . Here fMO gives the spatial and

temporal dependence of the laser pulse, HMO (A/m) gives the strength of the
magneto-optical field, and ± = ±1 its helicity. Thus for the Gaussian laser pulse from
the previous Tutorial, the spatial and temporal dependence of the magneto-optical
field is given by:

  | r  r0 |    (t  t 0 ) 2 
f MO r, t   exp  2  exp  2



 d / 4 ln( 2)   t R / 4 ln( 2) 

In Boris you can enable the magneto-optical effect with the moptical module
(addmodule meshname moptical).

The HMO parameter appears as a material parameter (see params), named Hmo.
Thus you can set the strength of the magneto-optical effect by setting the parameter
value. You can also set the fMO dependence as above by setting a material
parameter variation for Hmo (paramsvar). Thus for the above example you need to
set the spatial variation using a text equation set to:

exp(-sqrt((x/Lx - 0.5)^2 + (y/Ly - 0.5)^2) / ((d0/Lx)^2/(4*ln(2)))) * exp(-(t-

2*tau)^2/(tau^2/(4*ln(2))))

Exercise 29.1

Based on the simulation file from Exercise 28.1, study the effect of a train of
circularly polarised pulses with positive or negative helicities on skyrmion creation.
Use a reduced laser power density which does not raise the temperature above the
Curie temperature (set Q0 = 9e20 W/m3). Write a Python script to apply 20 laser
pulses with a 6 ps repetition period and strength of 40 MA/m. Compare the results
after 30 pulses for the two helicities.

142
Tutorial 30 – Spin-Wave Dispersion

Similar to the method of simulating frequency-swept FMR (see tutorial on FMR first),
we can simulate spin-wave dispersion by applying a sinc pulse which has not only a
time dependence, but also a spatial dependence. In the xy plane the excitation field
has the form:

H (t )  H e sinck c x  x0 sinck c  y  y0 sinc2f c t  t 0 

As before fc is the frequency cut-off (Hz) with t0 the temporal sinc pulse centre. To
excite spin-waves with non-zero wave-vector we also need a spatial dependence for
the sinc pulse. Here kc is the wave-vector cut-off (rad/m), and (x0, y0) is the spatial
sinc pulse centre. When obtaining the spin-wave dispersion, we need to consider the
direction of the wave-vector, k. Instead of sampling the average magnetisation, we
need to obtain a magentisation profile along a given direction. The direction in which
we sample is the wave-vector direction k. You can obtain a magentisation profile
using the dp_getexactprofile command. Also we need to sample the magnetisaiton
at fixed steps, determined by the Nyquist criterion (as for temporal samping):


S  (m)
kC

We only need to sample one component of the magnetisation, and the spin wave
dispersion is obtained by performing a 2D Fourier transform on the spatial-temporal
2D data, transforming it to wave-vector-frequency 2D space.

143
Consider a spin-wave waveguide as an elongated magnetic track. There are three
possible configurations, determined by the direction of the bias field.

1) Bias field along length – this is called the backward volume configuration.
2) Bias field along thickness – this is called the forward volume configuration.
3) Bias field along width – this is called the surface spin wave configuration.

In all three cases the excitation and analysed magnetisation component need to be
perpendicular to the bias field; for simplicity the analysed magnetisation component
can be chosen to be along the excitation direction. When simulating the spin-wave
dispersion it is important to choose the cell-size correctly, as the computed spin-
wave dispersion will be inaccurate at larger wave-vector values if the cellsize is not
small enough.

Exercise 30.1

Read through the reference IEEE Trans. Mag. 49, 524 (2013). Simulate the spin-
wave dispersion as described in this reference for the three spin-wave configurations
described above. Use periodic boundary conditions along the length only.

You should make the following modifications to the proposed problem:

1) Cellsize along length should be 1 nm; the specified 2 nm cellsize is too large
and results in a large discrepancy between computations and analytical
formulas at larger k values.
2) You should simulate for a total time of 2×t0 as explained in the tutorial on
FMR. The specified 5 ns simulation time for t 0 of 50 ps results in a very noisy
spectrum. For this exercise it is suggested you use a t 0 value of 200 ps,
although 100 ps is still fine. A value of 50 ps results in a coarse frequency
spectrum.

144
 3) The spatial samping interval should be 4 nm, not 2 nm, due to the Nyquist
criterion. A 2 nm spatial sampling interval results in wasted wave-vector
spectrum (try it!).

Hint: use an Hequation stage set to 'H0, He * sinc(kc*(x-Lx/2))*sinc(kc*(y-

Ly/2))*sinc(2*PI*fc*(t-t0)), 0' for the backward volume configuration.

When analysing the spin-wave dispersion you can use the following formula:

2A 2
w  wn  w M k (rad / s )
 0 M S2

Here wn is the resonance frequency (rad/s) for the nth spin-wave mode at k = 0, and
may be extracted from the computed spectrum at k = 0, or alternatively you can use
an analytical formula to predict it (not given here). The value wM is given as MS,
where  = µ0|e| (2.212761569×105 mA/s).

Figure 30.1 – Spin-wave dispersion computed in Exercise 30.1 together with

analytical predictions (even spin-wave modes excited n = 0, 2, 4, 6, 8, 10, 12). A
damping value of 0.01 was used.

145
Tutorial 31 – Two-Sublattice Model

In a micromagnetics formulation we can study antiferromagnetic, ferrimagnetic, as

well as binary ferromagnetic alloys using a two-sublattice model, where we consider
magnetic orders of two sub-lattices A, B, and couple them using inter-lattice
exchange stiffness terms. The two-sublattice stochastic LLB equation (sLLB) is given
below:

M i ~, i
 ~i M i  H eff , i  ~i M i  M i  H eff , i  H th, i 
t Mi
~||, i
i M i . H ||, i M i  ηth, i (i  A, B)
Mi

The reduced gyromagnetic ratio is given by ~i   i / 1   2 , i , and the reduced

transverse and longitudinal damping parameters by ~(||), i   (||), i / mi , where

mi (T )  M i (T ) / M S0, i , with M S0, i denoting the zero-temperature saturation

magnetisation, and Mi  |Mi|. The damping parameters are continuous at TN – the

phase transition temperature – and given by:
 T 
  , i   i 1  , T  TN
 3 i   ij me, j / me, i TN ~
 
 2T 
||, i   i  , T  TN
 3 i   ij me, j / me, i TN 
~
 
2T
 , i  ||, i  , T  TN
3TN

~
We denote TN the re-normalized transition temperature, given by:
~ 2TN
TN 
 A   B   A   B 2  4 AB BA

The micromagnetic parameters i and ij  [0, 1], are coupling parameters between
exchange integrals and the phase transition temperature, such that A + B = 1 and
|J| = 3kBTN. Here J is the exchange integral for intra-lattice (i = A,B) and inter-lattice

146
(i,j = A,B, i ≠ j) coupling respectively. For a simple antiferromagnet we have A = B =
AB = BA = 0.5. The normalised equilibrium magnetisation functions me,i are obtained
from the Curie-Weiss law as:


me, i  B me, i i  me, j ij 3TN / T  i 0 Hext / kBT ,
~


where B( x)  coth( x)  1 / x , and µi is the atomic magnetic moment. The magnetisation

length is not constant, and can differ from the equilibrium magnetisation length,
giving rise to a longitudinal relaxation field which includes both intra-lattice and inter-
lattice contributions:
 1  2 
3 k T  ~  2   m 2j  
1  mi   ij B N  ||, j 1  mi   e, j m 
m
H ||, i   ˆ .mˆ  1   m i , T  TN
~  2  ~  2  j  
 2 0  ||, i  me, i  2 0  i   ||, i  me, i  me, i
i 2
 me, j  
 1 3 ij k BTN  ~||, j me, j 
H ||, i   ~  ~  mˆ i .mˆ j  m i , T  TN
  0  ||, i  0  i   ||, i me, i  

ˆ i  mi / mi , and the relative longitudinal susceptibility is ~||, i   ||, i /  0 M S0, i , where:

Here m

k B T~||, i 
 ~
 ~
 i Bi 1  3 j TN B j / T   j 3 ij TN BiB j / T
 ~
 ~
  ~
1  3 i TN Bi / T 1  3 j TN B j / T   ij ji BiB j 3TN / T 2

and Bi  Bm e, i

m
e, i i
~

  me, j ij 3TN / T .

The direct exchange term includes the usual intra-lattice contribution, as well as
homogeneous and non-homogeneous inter-lattice contributions, and is given by:
2 Ai 4 Ah , i Anh, i
H ex, i   2M i  Mj   2M j
0 M e , i
2
0 M e , i M e , j 0 M e , i M e , j

The intra-lattice exchange stiffness Ai has the temperature dependence Ai  Ai0me2, i ,

whilst the inter-lattice exchange stiffnesses have the temperature dependences

Ah ( nh), i  Ah0( nh), i me, i me, j .

147
Finally, the terms Hth,i and ηth,i are stochastic quantities with zero spatial, vector
components, and inter-lattice correlations, and whose components follow Gaussian
distributions with zero mean and standard deviations given respectively by:

1 2k BT  , i   ||, i 
H thstd, i. 
 , i  i  0 M S0, iVt
2k BT ||, i  i M S0, i
 thstd, i. 
 0Vt

Here V is the stochastic computational cellsize volume, and Δt is the stochastic time-
step.

In Boris a mesh with the two-sublattice model may be added using the command:

addafmesh meshname rectangle

Default temperature dependences may be generated as for a ferromagnetic mesh

with the command (same command as for a ferromagnetic mesh, hence the naming
Curie):

curietemperature value (meshname)

Control of two-sublattice model meshes is exactly the same as for a ferromagnetic

mesh, but the list of parameters and available modules is different (see modules and
params). For a description of how the modules handle the two-sublattice model
meshes see the chapter on Modules in the manual.

Some important parameters you need to control are:

1) Micromagnetic  coupling factors. Set these using the tau command as:

tau tau11 tau22 tau12 tau21

2) Atomic moments µA, µB. Set these using the atomicmoment command as:

148
 atomicmoment mu1 mu2 meshname

Here mu1 and mu2 are the atomic moment values in units of the Bohr
magneton, and meshname must be the name of an antiferromagnetic mesh.
Note this command is also used to set the atomic moment for the LLB
equation for a ferromagnetic mesh, which only takes a single mu value – this
is the default behaviour, so giving the meshname parameter is important here.

3) Most magnetic parameters now have 2-sublattice values, so you can control
them separately if needed. The two inter-lattice exchange stiffness coupling
terms, Ah and Anh, only appear in two-sublattice model meshes.

When using the sLLB equation, by default the stochastic field generation time-step is
the same as the time-step using for the equation evaluation. You may need to
control this separately (set it to a larger value), and this is possible using the
setdtstoch command – see the output of the stochastic command for an interactive
display.

You can also set the stochastic cellsize value to be different than the magnetic
cellsize value using the scellsize command.

Exercise 31.1 (Advanced)

The two-sublattice sLLB equation produces a distribution of magnetisation lengths

on the two sub-lattices, mA, mB, which is expected to obey the following bi-variate
Boltzmann probability distribution:


Pi mA , mB   m exp 
2 M S0,iV 
 m2  m2
 i e,i    3 k T ~   m
2
i ij B N ||, j
2
j  me2, j 3 k T m 
2

 4i me,i k BT 2 ~||, i
i ij B N i
 me,i me, j  
(i, j  A, B, i  j )

149
Write a general-purpose Python script which tests the above equation against
computed bi-variate probability distributions for any possible combination of T, TN, i,
ij, µi, and M S0, i parameters. Test it for particular values (e.g. antiferromagnetic case).

Hint: there is a useful command built into Boris which computes a histogram for the
two-sublattice model, namely dp_histogram2 – see help for this command (there’s
also a dp_histogram command which works for ferromagnetic meshes). See Figure
31.1 for a typical output you should obtain from your script.

Figure 31.1 – Computed two-sublattice normalised magnetisation length probability

distribution at T/TN = 0.99 (colored surface) for an antiferromagnet in Exercise 31.1,
compared with the bi-variate Boltzmann probability distribution prediction (wire-
frame).

Exercise 31.2

Using the default antiferromagnetic mesh (addafmesh) in Boris, verify the Kittel
formula for antiferromagnetic resonance is reproduced (see F. Keefer and C. Kittel,
Phys. Rev. 85, 329 (1952)):

150
  H 0  H A 2 H E  H A 
w


Here HA is the uniaxial anisotropy field along the bias field, HA = 2K1/µ0MS, where MS
is the magnetisation length on a sub-lattice, and HE is the Weiss exchange field,
given by HE = 4Ah/µ0MS.

Hint: use the LLG equation, and adapt a previous exercise on frequency-swept FMR
to simulate a frequency-swept FMR peak and obtain the resonance frequency. You
should either use H0 set to zero, or small bias field values as the above formula
becomes inaccurate at large bias field values.

For the above exercise you should fit a Lorentz peak function with both symmetric
and asymmetric components for a more accurate result:
w  A( x  x0 )
f ( x)  y 0  S
4( x  x0 ) 2  w 2

Such a fitting procedure has already been built into Boris and can be accessed using
the dp_fitlorentz2 command.

Figure 31.2 – Antiferromagnetic resonance peak with fitted symmetric and

asymmetric Lorentz peak function, verifying Kittel’s formula in Exercise 31.2.

151
Tutorial 32 – Exchange Bias

The exchange bias field on a ferromagnetic layer from an antiferromagnet is given

as:

J
H ex  mj
0 M S t F

Here MS and tF are the saturation magnetisation and thickness of the ferromagnetic
layer, and mj is the exchange bias field direction from the antiferromagnet. This
effect may be modelled in Boris using the surfexchange module, since the bilinear
surface exchange field is given as:

J1
Hi  mj
0 M S t F

Here mj is the magnetisation direction on sub-lattice A of an interfacing

antiferromagnetic mesh. Thus in order to model exchange bias in Boris you need an
antiferromagnetic mesh (addafmesh), and a ferromagnetic mesh (addmesh) in
contact with it, and they both need the surfexchange module enabled, remembering
it is the top mesh (in order of z axis direction) which sets the J1 value.

Exercise 32.1

Simulate the exchange bias effect in a Fe 2nm thin film using a generic
antiferromagnetic material, 10 nm thick (use the addafmesh command to create a
default antiferromagnetic mesh). Enable in-plane uniaxial anisotropy for the
antiferromagnetic material (x axis), and set the antiferromagnetic sub-lattice A
magnetisation direction to result in a bias effect towards the +ve side.
You can add the Fe material from the materials database, and you should enable
cubic anisotropy for it. You can use periodic boundary conditions in the xy plane,
simulating an area of 320 × 320 nm2. Set the J1 surface exchange constant to 0.2
mJ/m2.

152
Tutorial 33 – Magneto-Elastic Effect

The magneto-elastic effect may be simulated in Boris using the melastic module.
The magneto-elastic effect can be included for a cubic crystal using a strain tensor.
The strain tensor is given as:

  xx  xy  xz 
 
S    xy  yy  yz  .
 
 xz  yz  zz 

Here we define the diagonal strain vector as Sd = (xx, yy, zz), and off-diagonal strain
vector as Sod = (yz, xz, xy). The strain tensor can have a spatial dependence, and
currently needs to either be loaded from ovf2 files (strain computed with an external
package), or alternatively a displacement vector field can be loaded (using ovf2 files,
computed externally), and the strain tensor computed as:

 u u y u z 
S d   x , , 
  x y z 
1  u y u z u x u z u x u y 
S od    ,  ,  
2  z y z x y x 

In the simplest case a uniform stress may be applied which results in a constant
strain with zero off-diagonal terms. In a future version an elastostatics solver as, well
as a dynamical elastic solver will be included.

From the strain tensor, for a cubic crystal with orthogonal axes e1, e2, e3, and
magneto-elastic constants B1, B2, we have the following diagonal and off-diagonal
energy density terms:

 m el,d  B1 (m.e1 ) 2 (S d .e1 )  (m.e 2 ) 2 (S d .e 2 )  (m.e3 ) 2 (S d .e3 )

 m el,od  2 B2 (m.e1 )(m.e 2 )(S od .e3 )  (m.e1 )(m.e3 )(S od .e 2 )  (m.e 2 )(m.e3 )(S d .e1 )

153
To set a uniform stress use the command (similar to the setfield command):

setstress magnitude polar azimuthal (meshname)

A uniform stress may also be set using the Sunif stage. When applying a uniform
stress, T, the strain tensor is generated based on the material Young’s modules and
Poisson ratio as:

 1   
1  
S    1   T
Ym 
   1 

Young’s moduls and Poisson ratio are available as material parameters as Ym and
Pr respectively. The magneto-elastic constants are available as material parameters
as MEc (2-component parameter for B1 and B2 respectively). The orthogonal axes
e1, e2, e3 are set by the magento-crystalline anisotropy axes (ea1 is e1, ea2 is e2 and
e3 = e1 × e2.

When applying a uniform stress the mesh origin (0, 0, 0) is a fixed point, and no
shear strain or physical displacement is allowed. This means a positive stress value
along the x axis results in elongation, whilst a negative stress value along the x axis
results in compression.

You can run computations with a non-uniform strain, but in the current version this
must be computed externally. There are two ways of setting a non-uniform strain.
The simplest method is to compute the displacement vector map u externally and
save it into a ovf2 file. This can then be loaded into the currently focused mesh using
the command (must have melastic module enabled):

loadovf2disp filename

After the displacement map is loaded the strain tensor used for computations is
obtained using the equations above.

154
You can also load the strain tensor directly, but this requires saving the diagonal and
off-diagonal components in two separate ovf2 files in vector data format. The
diagonal file will then contain vector data with strain components xx, yy, zz, and the
off-diagonal file will contain vector data with strain components yz, xz, xy (in this
order). The ovf2 files can then be loaded as:

loadovf2strain filename_diag filename_odiag

With the melastic module enabled, there is a separate cellsize for the strain tensor,
controlled using the mcellsize command, and this should be set before loading the
externally computed strain or displacement.

Exercise 33.1

Simulate hysteresis loops for a 5 nm thick Fe thin film (found in materials database)
with cubic anisotropy and melastic module enabled, along the x axis. Simulate three
cases: i) no strain, ii) compressive stress along the x axis of 100 MPa, iii) extensive
stress along the x axis of 100 MPa. Explain the differences between the 3 curves.

Figure 33.1 – Hysteresis loops computed in Exercise 33.1 for a Fe thin film, with and
without mechanical stress.

155
Tutorial 34 – Atomistic Modelling

This is a placeholder.

The current version has a simple cubic atomistic mesh (addameshcubic), which
implements a number of atomistic modules (Heisenberg exchange, DM and iDM
exchange, uniaxial and cubic anisotropies, Zeeman, dipole-dipole interaction, as well
as demagnetising fields obtained by computing magnetisation from atomic moments,
LLG and stochastic LLG; the heat and moptical modules are also enabled). Some
initial testing has already been done (e.g. computation of Curie temperature), but this
area of the software is set to be significantly expanded in the next version release
(bcc, fcc, hcp atomistic meshes, as well as full integration with micromagnetic
meshes (multi-layered demagnetisation and surface exchange) for multi-scale
computations; for this reason this area of the code has not been documented yet.

156
User-Defined Text Equations

Arbitrary text equations may be supplied by the user using fundamental functions, a
number of special functions, and mathematical operators as detailed below. These
equations are evaluated efficiently at run-time every iteration. Text equations use a
number of reserved variables depending on the context, including x, y, z to define
spatial variation, t to define temporal dependence, T to define temperature
dependence, and also use a number of reserved constants as listed below. Both
scalar and vector text equations may be supplied as appropriate. The contexts in
which text equations may be used are:

1. Setting stage values

The available stage types are: i) Hequation (set external field using a vector text
equation), ii) Vequation (set electrode potential drop using a scalar text equation), iii)
Iequation (set ground electrode current using a scalar text equation), iv) Tequation
(set mesh base temperature using a scalar text equation – applicable when heat
module disabled), v) Qequation (set heat source in heat equation using a scalar
equation).

Reserved variables:
x, y, z (spatial coordinates in meters, relative to mesh where used), t (stage time in
seconds).

Defined constants:
Lx, Ly, Lz (mesh dimensions in meters), Tb (mesh base temperature), Ss (stage
step).

Example: 0, 0, He * sinc(kc*(x-Lx/2))*sinc(kc*(y-Ly/2))*sinc(2*PI*fc*(t-t0)),
The above example is used with Hequation (vector equation), and applies a spatial
and temporal sinc pulse at the centre of the mesh with time centre at t0, for the z
field component. The constants He, kc, fc, t0 are defined by the user – see below.

157
2. Parameter temperature dependence

Any material parameter for which a temperature dependence is allowed (see output
of paramstemp command), may be assigned a temperature dependence defined
using a text equation. To set this either use the interactive console, or use the
following command:

setparamtempequation meshname paramname text_equation

Reserved variables:
T (temperature, either mesh base temperature if heat module not enabled, or the
local temperature if heat module is enabled – in the latter case the temperature can
be non-uniform and the parameter value is evaluated according to the local
computational cell temperature).

Defined constants:
Tb (mesh base temperature), Tc (set mesh Curie (or Néel) temperature).

Example: me(T/Tc)^2
The above example applies a squared Curie-Weiss scaling relation (me), for
temperatures ranging from 0 to Tc.

158
3. Parameter spatial variation

Any material parameter for which a spatial variation dependence is allowed (see
output of paramsvar command), may be assigned a spatial and temporal
dependence defined using a text equation. To set this either use the interactive
console, or use the following command:

setparamvar meshname paramname equation text_equation

Note in the above command the field “equation” specifies the type of generator used
to generate a spatial variation and must be inputted literally as above.

Reserved variables:
x, y, z (spatial coordinates in meters, relative to mesh where used), t (stage time in
seconds).

Defined constants:
Lx, Ly, Lz (mesh dimensions in meters).

Example: exp(-(x-Lx/2)^2/Sx) * exp(-(y-Ly/2)^2/Sy) / (exp(0)^2)

The above example applies a Gaussian spatial variation scaling in the xy plane at
the centre of the mesh.

159
Functions

Allowed functions in text equations are given below.

Function Name Description

sin Fundamental sin function.
sinc sinc(x) = sin(x) / x
cos Fundamental cos function.
tan tan(x) = sin(x) / cos(x)
sinh sinh(x) = [exp(x) - exp(-x)] / 2
cosh cosh(x) = [exp(x) + exp(-x)] / 2
tanh tanh(x) = sinh(x) / cosh(x)
sqrt Square root.
exp Fundamental exp function.
asin Inverse sin function.
acos Inverse cos function.
atan Inverse tan function.
asinh Inverse hyperbolic sin function.
acosh Inverse hyperbolic cos function.
atanh Inverse hyperbolic tan function.
ln Natural logarithm.
log Logarithm base 10.
abs Modulus function.
step Step function: step(x) = 0 for x < 0, = 1 for x>= 0.
swav Square wave function with period 2, s.t. swav(0+) = +1, swav(+) = -1.
twav Triangular wave function with period 2, s.t. twav(0) = +1, twav() = -1.
me Normalised Curie-Weiss law.
chi Normalised relative longitudinal susceptibility.
me1 Normalised Curie-Weiss law for sub-lattice A.
me2 Normalised Curie-Weiss law for sub-lattice B.
chi1 Normalised relative longitudinal susceptibility for sub-lattice A.
chi2 Normalised relative longitudinal susceptibility for sub-lattice B.
alpha1 Transverse damping temperature dependence.
alpha2 Longitudinal damping temperature dependence.

Any nested combination of functions is allowed.

160
Special equation expanders

1. sum

This is the sum function, with format given as: sum(i;low;high;func(<i>)).

Thus func(i) is summed for i ranging from integers low to high inclusive; note the
format i must appear in func, namely <i>. The variable named i can be changed to a
different string literal but must not clash with any equation variables, or reserved
names. For example the expression:
10
sum(j;0;10;sin(2*PI*<j>*x)), evaluates to:  sin 2jx 
j 0

Nested sum functions are also allowed, and may be combined with any of the
reserved function names in the table above. Note that using excessive limits (e.g.
integer range over 1000) in the sum function will result in very slow evaluation times
and should be avoided, especially if the evaluation is to be performed in every
computational mesh cell.

Allowed mathematical operators

Operator Symbol (in precedence order) Description

^ Exponentiation operator.
/ Division operator.
* Multiplication operator.
- Subtraction operator.
+ Addition operator.

All functions, variables, and constants must be linked by an operator, assumed

multiplication is not allowed, e.g. 2*sin(x) is a valid equation, 2sin(x) is not a valid
equation in the current program version.

Bracketing

Only round brackets are allowed, (, ), which must appear in equal numbers.

161
Numerical constants

Numbers may be specified in floating point or scientific format.

Reserved constants

A number of pre-defined numerical constants are available. Since the names in the
table below are reserved, they must not clash with any user defined constants.

Symbol Description
PI 3.1415926535897932384626433833
mu0 Free space permeability: 4*PI*10-7 (N/A2)
muB Bohr magneton: 9.27400968e-24 (Am2)
ec Electron charge: 1.60217662e-19 (C)
hbar Reduced Planck constant: 1.054571817e-34 (m2kg/s)
kB Boltzmann constant: 1.3806488e-23 (m2kg/s2K)
gamma Gyromagnetic ratio modulus: 2.212761569e5 (m/As)

User defined constants

Any number of user defined constants, named using alphanumeric strings, may be
given, with the restriction they must not clash with any of the reserved names in the
tables above, or equation variables. Equation constants may be given as:

equationconstants name value

Other related commands are delequationconstant, and clearequationconstants.

Vector and scalar equations

Scalar equation have a single component which must conform to the above rules.
Vector equations with 3 components are allowed where appropriate, and must be
specified using comma separators as: component1, component2, component3. Here
the three components are scalar equations.
162
Working with OVF2 Files

Data may be loaded and saved using the OVF2 file format introduced in OOMMF.
This allows setting magnetic shapes programmatically, as well as importing and
exporting data to other software which support this file format.

OVF2 files may also be used to save mesh data numerically, not only for
magnetisation, but for any mesh quantity which may be displayed on screen,
including material parameter spatial variation. The latter may also be set
programmatically using OVF2 files, via the ovf2 spatial variation generator (see
below).

OVF2 files allow scalar or vector formats, and both are handled in Boris. Data may
be saved in natural text format (text), single precision binary (bin4), or double
precision binary (bin8).

Commands which handle OVF2 files

loadovf2mag (renormalize_value) (directory/)filename

Load an OOMMF-style OVF 2.0 file containing magnetisation data, into the currently
focused mesh (which must be ferromagnetic), mapping the data to the current mesh
dimensions. By default the loaded data will not be renormalized: renormalize_value =
0. If a value is specified for renormalize_value, the loaded data will be renormalized
to it (e.g. this would be an Ms value).

saveovf2mag (n) (data_type) (directory\)filename

Save an OOMMF-style OVF 2.0 file containing magnetisation data from the currently
focused mesh (which must be ferromagnetic). You can normalize the data to Ms0
value by specifying the n flag (e.g. saveovf2mag n filename) - by default the data is
not normalized. You can specify the data type as data_type = bin4 (single precision 4

163
bytes per float), data_type = bin8 (double precision 8 bytes per float), or data_type =
text. By default bin8 is used.

saveovf2 (data_type) (directory/)filename

Save an OOMMF-style OVF 2.0 file containing data from the currently focused
mesh. You can specify the data type as data_type = bin4 (single precision 4 bytes
per float), data_type = bin8 (double precision 8 bytes per float), or data_type = text.
By default bin8 is used.

setparamvar meshname paramname ovf2 filename

Set the named parameter, paramname, spatial dependence for the named mesh,
meshname, using an ovf2 file with filename, located in current working directory. The
type of data in the OVF2 file must match the expected format of the parameter
(scalar or vector). Once you’ve loaded the ovf2 file you can display the set spatial
variation by selecting the ParamVar option under display, and clicking on the
required parameter under paramsvar.

The provided NetSocks.py module used for Python scripts, provides a convenient
method to handle OVF2 files: Write_OVF2, which allows writing a Python list into an
OVF2 file. An example of programmatically setting a magnetic shape using an OVF2
file generated in a Python script is given in Tutorial 0.

saveovf2param (data_type) (meshname) paramname (directory/)filename

Save an OOMMF-style OVF 2.0 file containing the named parameter spatial
variation data from the named mesh (currently focused mesh if not specified). You
can specify the data type as data_type = bin4 (single precision 4 bytes per float),
data_type = bin8 (double precision 8 bytes per float), or data_type = text. By default
bin8 is used.

164
Materials Database

Material definitions can be saved in a materials database. This includes base

material parameter values. The default database is called BorisMDB.txt. You can
see this using the command:

materialsdatabase

The default materials database can be updated from a shared database stored on a
server. The shared database can be seen at: https://boris-spintronics.uk/online-
materials-database. To update your local BorisMDB database using the latest
material parameter definitions, use the command:

updatemdb

You can also switch to an alternative custom database using the materialsdatabase
command. To add a new computational mesh with given material parameters you
can use the addmaterial command. The type of material will determine the type of
computational mesh generated. For example the ferromagnetic type will generate a
computational mesh with LLG/LLB solvers enabled. The conductor mesh type will
generate a computational mesh with only the transport and heat solvers enabled,
while the insulator mesh will only have the heat solver enabled (e.g. a substrate
material).

Users can also send in entries to be added to the centrally stored database. This can
be done using the requestmdbsync command. Before sending entries, you must
properly format the material entry. The procedure is described as follows.

1. Add a new entry to your local BorisMDB file.

Suppose you want to enter a new ferromagnetic material. First create a

ferromagnetic mesh in Boris (addmesh), and set as many parameter values
as possible. Next, add the entry to your local BorisMDB file using:

165
 addmdbentry meshname (materialname)

Here meshname is the name of the mesh as it appears in Boris (the one you
created using addmesh), and materialname is the name of the material, or
entry, you want to create, if different.

2. Edit the material description fields in BorisMDB.txt

There are 5 description fields for the entry:

Name, Formula, Type, Description, Contributor

Name is the name of your new material entry, which should not already be in
the shared online database. This should already be filled.

Formula is the symbolic formula for the material. Make sure to fill this.

Type is already filled for you, and is the type of computational mesh for which
the material applies.

Description should have a very brief description for the material entry, with
any useful information. Make sure to fill this.

Contributor is the name of the entry contributor; leave as N/A if you don’t want
to specify this.

There is another field called State. This specifies if the entry was taken from
the online database (SHARED) or if it’s a new user-created entry (LOCAL).
You don’t need to change this.

166
3. Set references for the parameters

After each parameter there is a column called DOI. This must hold a DOI
reference for where the material parameter value was taken from, or derived.
You can leave it as N/A only if not applicable, but in most cases should be
properly referenced.

4. Unspecified material parameters

If you cannot reasonably give an entry for a material parameter value then
override it as “N/A”.

5. Send in the entry

After properly formatting the entry, you can upload it to a holding database
using the requestmdbsync command:

requestmdbsync materialname (email)

Materialname is the name of the material you’ve just created. If you specify an
email address you will receive feedback about whether the entry was added
to the shared database or not – the entry will be verified for validity before
being added to the online materials database. Once entered there, it will be
visible at https://boris-spintronics.uk/online-materials-database, and other
users can update their databases with it.

167
Differential Equations

This section outlines the magnetization dynamics equations solved as selected using
the ode command. For descriptions of parameters used see the Material Parameters
section. To set an equation to solve and evaluation method use the setode
command as setode equation evaluation, e.g. setode LLG-STT RK4. In a Python
script you can set this as ns.setode(‘equation’, ‘evaluation’). For fixed time step
evaluation methods you can set the time step as setdt value. For a list of available
equations and evaluation methods see below.

A number of evaluation methods are available for the magnetization dynamics

equations. These are fixed step methods Euler (1st order), trapezoidal Euler (TEuler
– 2nd order) and Runge-Kuta (RK4 - 4th order). Adaptive time-step methods are the
adaptive Heun (AHeun – 2nd order), the multi-step Adams-Bashforth-Moulton (ABM –
2nd order), Runge-Kutta-Bogacki-Shampine (RK23 – 3rd order with embedded 2nd
order error estimator), Runge-Kutta-Fehlberg (RKF45 – 4th order with embedded 5th
order error estimator), Runge-Kutta-Cash-Karp (RKCK45 – 4th order with embedded
5th order error estimator), and Runge-Kutta-Dormand-Prince (RKDP54 – 5th order
with embedded 4th order error estimator). For static problems a steepest descent
solver is available, SDesc, using Barzilai-Borwein stepsize selection formulas.

Landau-Lifshitz-Gilbert (LLG)

The normalised LLG equation in implicit form is given by:

m m
 m  H  m 
t t

Here    0 g rel  e , where  e   g B /  is the electron gyromagnetic ratio and grel is a

relative g-factor (grel = 1 by default giving  = 2.212761569×105 m/As).

In explicit form the normalised LLG equation is given by:

168
m  
 mH  mm H
t 1  2
1  2

Two-Sublattice Landau-Lifshitz-Gilbert (LLG)

In antiferromagnetic (or ferrimagnetic, or binary ferromagnetic alloys) meshes a two-

sublattice model is used, where the two sub-lattices are labelled A, B, and each have
an LLG equation set. Coupling between the equations is done through the effective
fields, in particular the exchange field, but also the demagnetizing field. Each sub-
lattice has its own set of simulation parameters. In implicit and normalised form this
is given by:

m i m
  i m i  H i   i m i  i , (i  A, B)
t t

Landau-Lifshitz-Gilbert with Spin-Transfer Torques (LLG-STT)

The LLG equation can be complemented by Zhang-Li spin-transfer torques. In

implicit form this becomes:

m m
 γm  H  αm   u. m  βm  u.m
t t

The spin-drift velocity u is given by:

Pgμ B 1
uJ
2eM s 1  β 2

The LLG-STT equation in explicit form is given by:

169
m γ 
 mH  m  m  H  
t 1  2
1  2
1
1  βu.m  β   m  u.m   β   m.u.m m
1  2

Two-Sublattice Landau-Lifshitz-Gilbert with Spin-Transfer Torques (LLG-STT)

In implicit and normalised form this is given by:

m i m
 γ i m i  H i  α i m i  i  u i . m i  βm i  u i . m i , (i  A, B)
t t

Here:

Pgμ B 1
ui  J , (i  A, B)
2eM s ,i 1  β 2

Landau-Lifshitz-Bloch (LLB)

For non-zero temperature simulations the LLB equation should be used and in
implicit form is given by (un-normalised):

~ ~
M α M γα||
 γM  H  M  M.H M
t M t M

Here for T < TC (TC is the Curie temperature)     1 T / 3TC  ,  ||   2T / 3TC and

~    / m , ~||   || / m , where m is the magnetization length normalised to its zero

temperature value, i.e. m  | M | / M S0 . For T > TC    ||  2T / 3TC .

The effective field H must be complemented by a longitudinal susceptibility field

given by (which due to vector cross products only affects the longitudinal torque
term):

170
  m 2  m
1  2  , T  TC
 me  2 0 ~||
Hl  
 m , T  TC
  0 ~||

The field and temperature-dependent equilibrium magnetization, me, is given by:

 3T  H 
me (T )  B me C  0 ext  ,
 T k BT 

where B(x) = coth(x) – 1/x is the Langevin function, µ is the atomic moment.

The relative longitudinal susceptibility is given by (units 1/T), where || is the

longitudinal susceptibility (unitless):

 B( x)
~|| (T )    || (T ) /  0 M S0 , with x  me 3TC / T

k BT 1  B ( x)(3TC / T )

Further, we have the temperature dependences M S (T )  M S0 me (T ) , and

A(T )  A0 me2 (T ) for the exchange stiffness.

In explicit form the LLB equation becomes:

M γ ~
α γ 1 γ~
α||
 MH  M  M  H   M.H M
t 1 
2
1  M
2
M

171
Two-Sublattice Landau-Lifshitz-Bloch (LLB)

In explicit and un-normalised form this is given by:

M i ~, i ~||, i
~ ~
  i M i  H eff , i   i M i  M i  H eff , i    i M i . H ||, i M i , (i  A, B)
t Mi Mi

Here ~i   i / 1   2 , i  and Mi  |Mi|. As before ~(||), i   (||), i / mi , where mi (T )  M i (T ) / M S0, i

with M S0, i denoting the zero-temperature saturation magnetisation. The damping

parameters are continuous at TN – the phase transition temperature – and given by:

 T 
  , i   i 1  , T  TN
 3 i   ij me, j / me, i T~N 
 
 2T 
 ||, i   i   , T  TN
 3 i   ij me, j / me, i T~N 
 
2T
  , i   ||, i  , T  TN
3TN

~
Here we denote TN the re-normalized transition temperature, given by:

~ 2TN
TN 
 A   B   A   B 2  4 AB BA

The micromagnetic parameters i and ij  [0, 1], are coupling parameters between
exchange constants and the phase transition temperature, such that A + B = 1 and
|J| = 3kBTN. Here J is the exchange constant for intra-lattice (i = A,B) and inter-
lattice (i,j = A,B, i ≠ j) coupling respectively.

172
The normalised equilibrium magnetisation functions me,i are obtained from the Curie-
Weiss law as:


me, i  B me, i i  me, j ij 3TN / T  i 0 H ext / k BT
~


The longitudinal relaxation field includes both intra-lattice and inter-lattice

contributions as:

 1  m 2  3 ij k BTN  ~||, j  m 2  me, j  m 2j  

H ||, i   
~ 1  2  
i  
 ~ 1 2 

i 

m i .m j 1  2
ˆ ˆ   m i , T  TN

 2 0  ||, i  me, i  2 0  i   ||, i  me, i  me, i  me , j  
 1 3 ij k BTN  ~||, j me, j 
H ||, i   ~  ~  mˆ i .mˆ j m i , T  TN
  0  ||, i  0  i   ||, i me, i  

ˆ i  mi / mi , and the relative longitudinal susceptibility is ~||, i   ||, i /  0 M S0, i , where:

Here m

k BT~||, i 
 ~
 ~
 i Bi 1  3 j TN Bj / T   j 3 ijTN BiBj / T
 ~
 ~
  ~
1  3 iTN Bi / T 1  3 jTN Bj / T   ij ji BiBj 3TN / T2 ,

 
and Bi  Bm e , i me, i i  me, j ij 3TN / T .
~

173
Landau-Lifshitz-Bloch with Spin-Transfer Torques (LLB-STT)

In implicit form we have the LLB-STT equation as:

~ ~
M α M γα||
 γM  H  M  M.H M  u.M  β M  u.M
t M t M M

In this case the spin-drift velocity is given by:

Pgμ B 1
uJ
2eM S 1  β 2

In explicit form the LLB-STT equation becomes:

M γ ~α γ 1 γ~α||
 MH  
M  M  H   M.H M 
t 1  α 2 1  α 2 M M

1  ~
   α β    
β  ~α   M  u. M  ~α  β  ~α   M.u. M M 
 
1  α 2 

1  u. M
M M
2


Two-Sublattice Landau-Lifshitz-Bloch with Spin-Transfer Torques (LLB-STT)

In explicit and un-normalised form this is given by:

M i γi ~α γ γ i ~α||,i
M i  M i  H i   M i .H i M i 
 ,i i 1
 Mi  Hi 
t 1  α 2,i 1  α 2,i M i Mi

1  ~
   α β    
β  ~α,i  M  u .M  ~α,i β  ~α,i  M .u .M M 

1  α 2,i 

1  ,i i u i . M i
Mi
i i i
Mi
2 i i i i


Here:

Pgμ B 1
ui  J
2eM S ,i 1  β 2

174
Stochastic Landau-Lifshitz-Gilbert (sLLG)

The explicit and normalised sLLG equation is given as:

m  
 m  H  H thermal  m  m  H  H thermal
t 1  2
1  2

Each vector component of the thermal field follows a Gaussian distribution with zero
mean and standard deviation given by:

2k BT
H 
 0 M s0Vt

V is the volume of the stochastic computational cell, and t is the time step used to
update the stochastic field. The stochastic field has zero spatial and vector
component correlations.

Two-Sublattice Stochastic Landau-Lifshitz-Gilbert (sLLG)

In explicit and normalised form this is given by:

m i  
  i 2 m i  H i  H th, i   i i 2 m i  m i  H i  H th, i , (i  A, B)
t 1 i 1 i

As with the sLLG equation the thermal field standard deviation is given by:

2 i k B T
H thstd, i. 
 i  0 M S0, iVt

175
Stochastic Landau-Lifshitz-Bloch (sLLB)

For the stochastic LLB equation we have both a thermal field and thermal torque,
and is given by:

M γ ~
α γ 1 γ~
α||
  ~2 M  H  ~2 M  M  H  H therm al  M.H M  ηtherm al
t 1  α 1  α M M

The components of the thermal field and torque follow Gaussian distributions with no
correlations, zero mean and standard deviation given respectively by:

1 2k BT (    || )
H 
  0 M s0Vt

2k BT ||M s0
 
 0Vt

Two-Sublattice Stochastic Landau-Lifshitz-Bloch (sLLB)

This is given by:

M i ~, i ~||, i
~ ~
  i M i  H eff , i   i M i  M i  H eff , i  H th, i    i M i . H||, i M i  ηth, i (i  A, B)
t Mi Mi

As with sLLB we have (i = A, B):

1 2k BT  , i   ||, i 
H thstd, i. 
 , i  i  0 M S0, iVt
2k BT ||, i  i M S0, i
 std .

 0Vt
th , i

176
Stochastic Landau-Lifshitz-Gilbert with Spin-Transfer Torques (sLLG-STT)

This is similar to the LLG-STT equation, but also has the thermal field from the sLLG
equation added.

Two-Sublattice Stochastic Landau-Lifshitz-Gilbert with Spin-Transfer Torques (sLLG-

STT)

This is similar to the two-sublattice LLG-STT equation, but also has the thermal field
from the sLLG equation added.

Stochastic Landau-Lifshitz-Bloch with Spin-Transfer Torques (sLLB-STT)

This is similar to the LLB-STT equation, but also has the thermal field from the sLLB
equation added to the damping torque term, as well as the additional thermal torque
term.

Two-Sublattice Stochastic Landau-Lifshitz-Bloch with Spin-Transfer Torques (sLLB-

STT)

This is similar to the two-sublattice LLB-STT equation, but also has the thermal field
from the two-sublattice sLLB equation added to the damping torque term, as well as
the additional thermal torque term.

177
Equations with Spin Accumulation

The LLG, LLB, sLLG, and sLLB equations also appear in the forms LLG-SA, LLB-
SA, sLLG-SA, and sLLB-SA. When using these equations the spin transport solver is
enabled and a spin accumulation S is calculated. This gives rise to bulk and
interfacial torques which are added to the respective equation. For example for the
LLG equation we obtain the LLG-SA equation as:

M  M
 M  H  M  TS
t M t

The bulk spin-accumulation torque is given by:

m  S  2e m  m  S 
De D
TS  
J
2


This is included as an additional effective field in the explicit forms of the equations:

De  S m  S 
HS   2
 | M |  2J  

Note there are no SA version for the STT equations. This is because the Zhang-Li
STTs result from the bulk TS torque as a special case (see e.g. S. Lepadatu,
Scientific Reports 7, 12937 (2017)).

Interfacial spin-accumulation torques are also present when N/F interfaces are used:

TSint erface 
g B
ed h
     
Re G  m  m  VS   Im G  m  VS ,

where VS = VS,F – VS,N and VS  De /  e /  B S .

178
This is included as an additional effective field in the explicit forms of the equations:

HS 
 1 g B
  
 | M | ed h
  
Re G  m  VS  Im G  VS

Currently antiferromagnetic meshes do not have a drift-diffusion model included, so

the SA versions are not active with two-sublattice equations. This will be included in
a future version.

Static Landau-Lifshitz-Gilbert (LLGStatic)

This equation is used for static problems, i.e. where only the relaxed magnetization
state is required. It is the explicit LLG equation without the precession term, and with
the damping factor set to 1. This is given by:

m 
  mmH
t 2

179
Modules

To add a module to simulations use the addmodule command as addmodule

meshname modulename, e.g. addmodule permalloy iDMexchange. In Python
scripts you can use ns.addmodule(‘meshname’, ‘modulename’). To add a
supermesh module you need to use the meshname as supermesh, e.g. addmodule
supermesh sdemag. To remove a module from simulations use the delmodule
command, with the same parameters as addmodule. For a list of available module
names see below. To see a list of available modules in the console use the modules
command.

Modules typically correspond to an additive field in the total effective field H

appearing in the equations shown in the Differential Equations section:

H  H eff  H1  H 2  ...

Most modules also have an energy density term associated with their effective field
contributions, available as an output data parameter.

All contributions are evaluated on a cell-centered uniform finite difference mesh, with
all differential operators evaluated to second order accuracy.

aniuni – Uniaxial Magneto-Crystalline Anisotropy

Effective field contribution:

2 K1 4K 2
H (m.e A )e A  [1  (m.e A ) 2 ](m.e A )e A
0 M S 0 M S

Energy density term (output data parameter: e_anis):

  
  K1 1  m.e A 2  K 2 1  m.e A 2 
2

180
For two-sublattice models the same expression is used, but K1, K2, and MS can
have different values on the two sublattices.

anicubi – Cubic Magneto-Crystalline Anisotropy

Effective field contribution:

2 K1
H [e1 (  2   2 )  e 2  ( 2   2 )  e 3 ( 2   2 )]
0 M S
2K 2
 [e1 2 2  e 2 2  2  e 3 2  2 ]
0 M S

Here  = m.e1,  = m.e2, and  = m.e3, where e3 = e1 × e2.

Energy density term (output data parameter: e_anis):

  K1[ 2  2   2 2   2 2 ]  K 2 2  2 2

For two-sublattice models the same expression is used, but K1, K2, and M S can
have different values on the two sublattices.

demag_N – Stoner-Wohlfarth Magnetostatic Interaction

Effective field contribution:

H i   N i M i (i  x, y, z )

Here Nz = 1 - Nx - Ny.

Energy density term (output data parameter: e_demag):

0
  M.H
2

181
For two-sublattice models the same expression is used, but applied to the average
magnetisation value.

demag - Magnetostatic Interaction

Effective field contribution:

H (r0 )    N(r  r0 )M (r )dr

rV

Here N is a rank-2 tensor with the following symmetry:

 N xx N xy N xz 
 
N   N xy N yy N yz 
N N yz N zz 
 xz

N is computed using the formulas in A.J. Newell et al., “A Generalization of the

Demagnetizing Tensor for Nonuniform Magnetization” J. Geophys. Res. 98, 9551
(1993).

Energy density term (output data parameter: e_demag):

0
  M.H
2

The convolution function is evaluated using the convolution theorem, i.e. both N and
M are transformed using an FFT algorithm, multiplied in the transform space, then H
is obtained using the inverse FFT; M is zero-padded before computing the FFT.

For two-sublattice models the same expression is used, but applied to the average
magnetisation value.

182
DMExchange – Dzyaloshinskii-Moriya Bulk Exchange Interaction

Effective field contribution:

2D
H M
 0 M S2

Non-homogeneous Neumann boundary conditions are used to evaluate the curl

operator (single lattice only):

M D
 nM
n 2 A

The DM exchange field adds to the direct exchange field.

Energy density term (output data parameter: e_exch):

0
  M.H
2

For two-sublattice models the same expression is used, but D and M S can have
different values on the two sublattices.

exchange – Direct Exchange Interaction

Effective field contribution:

2A
H  2M
0 M S
2

Homogeneous Neumann boundary conditions are used to evaluate the Laplacian

operator.

183
Energy density term (output data parameter: e_exch):

0
  M.H
2

This is equivalent to:

 m  2  m  2  m  2 
  A        
 x   y   z  

For two-sublattice models the same expression is used, but A and MS can have
different values on the two sublattices. Moreover for two-sublattices there are
additional inter-lattice exchange interactions, which include both homogeneous and
non-homogeneous contributions. In this case the full exchange field is given by:

2 Ai 4 Ah, i Anh, i
H ex, i   2M i  Mj   2 M j , (i, j  A, B, i  j )
0 M e, i
2
 0 M e, i M e, j  0 M e, i M e, j

heat – Heat Equation Solver

The heat equation in the 1-temperature model with Joule heating and any other
additional heat sources (S) is given by:

T (r, t ) J2
Cρ  .KT (r, t )   S
t 

Robin boundary conditions are used to evaluate the differential operators.

The heat equation is evaluated using the simple forward-time centered-space

method. The heat equation time-step required is normally comparable to the
magnetization equation time-step thus a more time-efficient method (e.g. Crank-
Nicolson) is not normally required.

184
In the two-temperature model, the heat equation is given as:

Te (r, t )
Ce ρ  .KTe (r, t )  Ge (Te  Tl )  S
t
T (r, t )
Cl ρ l  Ge (Te  Tl )
t

Here Ce and Cl are the electron and lattice specific heat capacities,  is the mass
density, K is the thermal conductivity, and Ge is the electron-lattice coupling
constant, typically of the order 1018 W/m3K.

iDMExchange – Dzyaloshinskii-Moriya Interfacial Exchange Interaction

Effective field contribution for thin film in xy plane:

2D  M z M z M x M y 
H  , ,  
 0 M S2  x y x y 

Non-homogeneous Neumann boundary conditions are used to evaluate the

differential operators:

M D
 (zˆ  n)  M
n 2A

The iDM exchange field adds to the direct exchange field.

Energy density term (output data parameter: e_exch):

0
  M.H
2

For two-sublattice models the same expression is used, but D and M S can have
different values on the two sublattices. Also the boundary conditions for differential

185
operators are modified due to the non-homogeneous inter-lattice exchange coupling
term, and given by:

M i
n

Di
2 Ai (1  ci2 )
 
(zˆ  n)  M i  ci M j  , (i, j  A, B, i  j ) ,

where ci = Anh / 2Ai.

melastic – Magneto-elastic effect

The magneto-elastic effect can be included for a cubic crystal using a strain tensor.
The strain tensor is given as:

  xx  xy  xz 
 
S    xy  yy  yz  .
 
 xz  yz  zz 

Here we define the diagonal strain vector as Sd = (xx, yy, zz), and off-diagonal strain
vector as Sod = (yz, xz, xy). The strain tensor can have a spatial dependence, and
currently needs to either be loaded from ovf2 files (strain computed with an external
package), or alternatively a displacement vector field can be loaded (using ovf2 files,
computed externally), and the strain tensor computed as:

 u u y u z 
S d   x , , 
 x y z 
1  u y u z u x u z u x u y 
S od    ,  ,  
2  z y z x y x 

In the simplest case a uniform stress may be applied which results in a constant
strain with zero off-diagonal terms. In a future version an elastostatics solver as, well
as a dynamical elastic solver will be included.

186
From the strain tensor, for a cubic crystal with orthogonal axes e1, e2, e3, and
magneto-elastic constants B1, B2, we have the following diagonal and off-diagonal
energy density terms:

 m el,d  B1 (m.e1 ) 2 (S d .e1 )  (m.e 2 ) 2 (S d .e 2 )  (m.e3 ) 2 (S d .e3 )

 m el,od  2 B2 (m.e1 )(m.e 2 )(S od .e3 )  (m.e1 )(m.e3 )(S od .e 2 )  (m.e 2 )(m.e3 )(S d .e1 )

1  m el
The effective field can be computed using the usual formula: H m el   .
 0 M S m
Currently this module is not enabled for two-sublattice models.

moptical – Magneto-optical effect

This module applies a z-axis field as given by the Hmo parameter (A/m) as:

H MO    H MO
0
f MO r, t zˆ

Here   H MO
0
is set using the Hmo parameter, and fMO is its spatial (and temporal)

variation.

For two-sublattice models the same field is applied to both sublattices.

Oersted – Oersted Field

Effective field contribution:

H(r0 )   K (r  r )J
rV
0 C (r )dr

Here K is a rank-2 tensor with the following symmetry:

187
  0 K xy K xz 
 
K    K xy 0 K yz 
 K  K yz 0 
 xz

K is computed using the formulas in B. Krüger, “Current-Driven Magnetization

Dynamics: Analytical Modeling and Numerical Simulation”, PhD Dissertation,
University of Hamburg (2011) – Appendix D, page 118.

For two-sublattice models the Oersted field is applied equally to both sublattices.

roughness – Roughness Field and Staircase Magnetostatic Corrections

Effective field contribution computed on the coarse mesh (i.e. the actual mesh
discretisation used at run-time with NV number of discretisation cells):

 
H (r0 )    N(r  r0 )G (r, r0 ) M (r0 ) (r0  V )
 rV 

Here N is the demagnetizing tensor computed on the fine mesh with NVr number of
discretisation cells, and:

 NV
  1 r  r0  VR
G(r, r0 )   NVr
  1 r  r0  V  VR

V is the smooth body without roughness and VR is the mesh with roughness, and we
require VR  V. If the coarse cellsize has dimensions (hx, hy, hz), the fine cellsize
must have dimensions (hx / mx, hy / my, hz / mz), where the m factors are integers.
The function  N(r  r )G(r, r )
rV
0 0 is computed at initialisation on the finely discretised

mesh then averaged up to the coarse mesh (each coarse cellsize value is obtained
as an average of its contained fine cellsize values).

Energy density term (output data parameter: e_rough):

188
 0
  M.H
2

Details can be found in: S. Lepadatu, “Effective field model of roughness in magnetic
nano-structures” J. Appl. Phys. 118, 243908 (2015).

For two-sublattice models the roughness field is computed using the average
magnetization, and applied equally to both sublattices.

sdemag – Supermesh Magnetostatic Interaction

I. Supermesh demagnetization (multiconvolution 0).

The same formulas as for the Demag module are used when computing
demagnetizing fields on the uniformly discretised super-mesh. The ferromagnetic
super-mesh may have a cellsize which differs from that of the individual
ferromagnetic meshes. In this case a weighted average smoother is used to transfer
magnetization to the super-mesh and demagnetizing field values back from the
super-mesh.

Consider a discrete distribution of magnetization values M at points V = {ri; iP}. Let

h be the cellsize of the input mesh, with the set of cells {ci; iP} centered around the
points ri. To obtain the magnetization value at a point r in a cell c with dimensions h
~
we introduce the definitions di = |rri|, dV = |h+h|/2, and d i  dV  d i . The weighted

average is given as:

M(r)   wi M(ri )
iP

where

189
 ~
d i i
wi  ~
dT
1, ci  c  
i  
0, otherwise
~ ~
d T   d i i
iP

II. Multi-Layered convolution (multiconvolution 1)

A generalisation of the single layer convolution algorithm is used here. We can write
the convolution sum as:

H(rkl )    N(r  r , h , h )M(r ),

i 1,...,n
kl ij k i ij k  1,..., n; rkl  Vk
rij Vi

In the demagnetizing tensor for the equation above we explicitly specify the cellsize,
h, of the two computational meshes the tensor relates. Since we have n terms of the
form appearing in the single layer convolution sum, we can again apply the
convolution theorem. This time for each output mesh (H) we have n input meshes
(M), together with n kernels. Thus to calculate the outputs in all n meshes we require
a total of n2 sets of kernel multiplications in the transform space. This is illustrated in
the figure below.

Multi-Layered convolution algorithm for n computational meshes. The

magnetization input of each mesh is transformed separately using a FFT algorithm,
either directly (dotted line), or by first transferring to a scratch space with a common

190
discretisation cellsize, using a weighted average smoother (solid lines). In the
transform space the inputs are multiplied with pre-computed kernels for a total of n2
sets of point-by-point multiplications. Finally the output demagnetizing fields are
obtained using an inverse FFT algorithm, which are set directly in the output meshes
(dotted line), or transferred using a weighted average smoother if the discretisation
cellsizes differ (solid lines).

SOTfield – Spin-Orbit Torque Field

The spin-orbit torque is given by:

B Jc
TSOT   SHA, eff (m  m  p   rG m  p)
e dh

Here p = z × eJc. This results in an effective field in the magnetization dynamics

equation given by:

 J
 SHA, eff B c m  p  rG p 
1
H SOT  
M S e dh

For two-sublattice models the same expression is used, but  and MS can have
different values on the two sublattices.

strayfield – Stray field from magnetic dipoles

If Md is the magnetization of a uniformly magnetized prism with dimensions d, then

the magnetic field (stray field) at a distance r from the centre of the prism is given by:
H  N d (r, d)M d

Here Nd is a rank-2 tensor with the following symmetry:

191
  N xx N xy N xz 
 
N d   N xy N yy N yz 
N N yz N zz 
 xz

Nd is computed using the formulas in A. Andreev et al., “Universal Method for the
Calculation of Magnetic Microelectronic Components: the Saturated Ferromagnetic
Rectangular Prism and the Rectangular Coil.” ICSE2000 Proceedings, Nov. 2000,
187. Note these formulas are equivalent to the Newell formulas used to compute the
demagnetizing tensor.
For two-sublattice models the field is applied equally to both sublattices.

surfexchange – Surface Exchange Interaction

Let mi and mj be the normalised magnetization values of two cells, i and j, which are
surface exchange coupled across a gap between two ferromagnetic meshes. Let 
be the thickness of the ferromagnetic layer for which the surface exchange field is
computed. The surface exchange field at cell i, from cell j, is given as:

J1 2J 2
Hi  mj  (m .m )m
0 M S  0 M S  i j j

The surface exchange field is applied for all cells along the z direction if a 3D
simulation mesh is used (surface exchange field applicable for thin films).

Energy density term (output data parameter: e_surfexch):

J1 J
  m i .m j  2 (m i .m j ) 2
 

The surfexchange module also allows coupling between an antiferromagnetic mesh

and a ferromagnetic mesh, with the resultant coupling being the exchange bias field.
In this case the same formulas above are used, but the coupling is only done to/from
sub-lattice A.

192
transport – Charge and Spin-Transport Solver

Charge Transport

When solving only for the charge current density, a Poisson-type equation for V is
solved as:

 2V  
V .  


For V Dirichlet boundary conditions are used at boundaries containing a fixed

potential electrode, otherwise Neumann boundary conditions are used. The
conductivity may have an AMR contribution (AMR given as a percentage value)
calculated as:

0
 ,
1  ( AMR / 100)d 2

where

JC . M
d .
| J C || M |

From V the charge current density is obtained as Jc = -V, and is used for Joule
heating computations and to obtain spin torques (Zhang-Li STT and SOT).

The Poisson equation is evaluated using the successive over-relaxation (SOR)

algorithm with black-red ordering for parallelization.

Charge and Spin Transport

Charge and spin current densities are given as (see S. Lepadatu, “Unified treatment
of spin torques using a coupled magnetization dynamics and three-dimensional spin
current solver” Scientific Reports 7, 12937 (2017) for details):

193
 S m   SHADe e   S  P  E  P  2  E  B
2
e
J C  E   D De
B B 2e en

B     B 2
JS   PE  m  DeS   SHA B εE  B2  ei  m  im  z  E   xm   ym
e e 2e i e3n

Where:

Ei  m
  im.m

B  z  xm   y m .m

Here E and B are the directions of the emergent electric field due to charge
pumping, and emergent magnetic field due to topological Hall effect respectively.

With the full spin transport solver enabled both V and S are computed using
Poisson-type equations as:

 2V  
V .     D De e
.S m
  B


P
2e

 xm
   xm   y m
   ym  m 
   2xm   2y m .m 
P
  
 2 x  2xym   y m   xm   2y m .m  y  2xm   y m   xm   2xym .m .V
en
 

and

 B P
 2S   E. m   B P  2V m   SHA  B .εE
e De e De De e


 B
2
2e De

 xm
   xm   ym
   ym  m 
   2xm   2y m 
 B 2
 3
e nDe
   
Ex  2xym   y m   xm   2y m  E y  2xm   y m   xm   2xym 
S S  m m  S  m 
  2 
2sf J 2

194
For boundaries containing an electrode with a fixed potential, differential operators
applied to V use a Dirichlet boundary condition. For other external boundaries the
following non-homogeneous Neumann boundary conditions are used:

V .n   SHA
De e
  S .n
 B
 B
S .n  SHA εE.n
De e

At N/F composite media boundaries the following conditions are applied:

  
J C .n N  J C .n F   G   G  V  G   G  VS .m 
J S .n N  J S .n F    
2 B
e
 
Re G  m  m  VS   Im G  m  VS 
J S .n F 

e
 G  G V .mm  G
B  
S


 G  Vm 

Spin pumping is included on the N side of the above equations as:

J Spum p 
B 
2 
e 
 
Re G  m 
m
t
 Im G 
m 
t 
 

At N/F interfaces, interfacial spin torques are obtained as (hF is the discretisation
cellsize of the F layer in the direction normal to the composite media boundary) :

TS 
g B
ehF
  
Re G  m  m  VS   Im G  m  VS   

From S, bulk spin torques are obtained as:

m  m  S 
De De
TS   mS 
2
J 2

195
Both Poisson equations are evaluated using the SOR algorithm with black-red
ordering for parallelization. Note, whilst the SOR algorithm is robust in evaluating the
spin transport equations in arbitrary multi-layers with composite media boundary
conditions, it does suffer from slow convergence in particular for lower target solver
errors. This algorithm is due to be replaced with a more efficient method in the next
version. Currently the alternating direction implicit method with parallelized Thomas
algorithm, as well as a FFT-based Poisson solver are being evaluated. Another
possibility is a bi-conjugate gradient method.

Zeeman – Applied Magnetic Field

Effective field contribution:

H  H ext

Energy density term (output data parameter: e_zee):

  0 M.H

For two-sublattice models the field is applied equally to both sublattices.

196
Material Parameters

To set a material parameter value use the setparam command as setparam

meshname paramname value, e.g. setparam permalloy A 1e-11, or even setparam
permalloy A 10pJ/m. In Python scripts you can set a parameter value as
ns.setparam(‘meshname’, ‘paramname’, value). For list of available parameter
names see below. To see a list of available parameters in the console use the
params command.

Format:
paramname: name in equations (units)
Description.

A: A (J/m)
Exchange stiffness.

A_AFM: Ai (J/m)
Exchange stiffness.

Ah: Ah,i (J/m3)

Homogeneous inter-lattice exchange coupling per lattice constant.

Anh: Anh,i (J/m3)

Non-homogeneous inter-lattice exchange stiffness.

amr: AMR (%)

Anisotropic magneto-resistance as a percentage of base resistance.

beta:  (unitless)
Spin-transfer torque non-adiabaticity parameter.

betaD: D (unitless)
Diffusion spin polarisation.

197
cHA: cHA (unitless)
Applied field spatial variation parameter, which multiplies the applied field value.

cpump_eff: cpump_eff (unitless)

Charge-pumping efficiency.

cT: cT (unitless)
Set temperature spatial variation parameter, which multiplies the set temperature
value. To enable spatial variation of temperature you need to have the heat module
enabled. If you want the set temperature to remain constant you need to disable the
heat equation by using setheatdt 0.

D: D (J/m2)
Dzyaloshinskii-Moriya exchange constant.

D_AFM: Di (J/m2)
Dzyaloshinskii-Moriya exchange constant.

damping:  (unitless)
Gilbert magnetization damping.

damping_AFM: i (unitless)
Gilbert magnetization damping.

De: De (m2/s)
Electron diffusion constant.

density:  (kg/m^3)
Mass density.

ea1: ea1 (unit vector)

Uniaxial magneto-crystalline anisotropy symmetry axis, or first cubic magneto-
crystalline anisotropy symmetry axis.

198
ea2: ea2 (unit vector)
Second cubic magneto-crystalline anisotropy symmetry axis (ea3 = ea1 × ea2).

elC:  (S/m)
Base electrical conductivity.
flSOT: rG (unitless)
Field-like spin orbit torque coefficient.

G_e: Ge (W/m3K)
Electron-lattice coupling constant (two temperature model).

Gi: G, G (S/m^2)

Interface spin-dependent conductivity (for majority and minority carriers). The top
contacting mesh sets the interface value, thus Gi is available in both magnetic and
non-magnetic meshes.

Gmix: G = Re{G} + i Im{G} (S/m^2)

interface spin-mixing conductivity (real and imaginary parts). The top contacting
mesh sets the interface value, thus Gmix is available in both magnetic and non-
magnetic meshes.

grel: grel (unitless)

Relative electron gyromagnetic ratio.

grel_AFM: grel,i (unitless)

Relative electron gyromagnetic ratio.

Hmo: Hmo (A/m)

Magneto-optical field strength.

iSHA: SHA (unitless)

Spin Hall angle used for the inverse spin Hall effect.

199
J1: J1 (J/m2)
Bilinear surface exchange coupling. For coupled meshes it is the top mesh that sets
the J values.

J2: J2 (J/m2)
Biquadratic surface exchange coupling. For coupled meshes it is the top mesh that
sets the J values.

K1: K1 (J/m3)
Magneto-crystalline anisotropy energy.

K1_AFM: K1i (J/m3)

Magneto-crystalline anisotropy energy.

K2: K2 (J/m3)
Magneto-crystalline anisotropy energy, higher order.

K2_AFM: K2i (J/m3)

Magneto-crystalline anisotropy energy, higher order.

l_J: J (m)
Spin exchange rotation length.

l_phi:  (m)
Spin dephasing length.

l_sf: sf (m)

Spin-flip length.

MEc: B1, B2 (J/m3)

Magneto-elastic coefficients.

200
Ms: Ms (A/m)
Saturation magnetization.

Ms_AFM: Msi (A/m)

Saturation magnetization.
n: n (m-3)
Conduction electrons density.

Nx, Ny: Nxy (unitless)

In-plane demagnetizing factors (used by demag_N module)

P: P or  (unitless)
Charge current spin polarization.

Pr:  (unitless)
Poisson’s ratio.

pump_eff: pump (unitless)

Spin pumping efficiency.

Q: Q (W/m3)
Heat source added to the heat equation. Can be non-uniform by setting a spatial
variation.

SHA: SHA (unitless)

Spin Hall angle used for the spin Hall effect.

shc: C (J/kgK).
Specific heat capacity (total – one temperature model, lattice – two temperature
model).

shc_e: C (J/kgK).
Electronic specific heat capacity (two temperature model).

201
susrel: || (As2/kg)
Longitudinal (parallel) susceptibility divided by µ0Ms.

susrel_AFM: ||,i (As2/kg)

Longitudinal (parallel) susceptibility divided by µ0Ms.

tau_ii: ii (unitless)

Coupling between exchange constants and phase transition temperature: two-
sublattice model, intra-lattice.

tau_ij: ij (unitless)

Coupling between exchange constants and phase transition temperature: two-
sublattice model, inter-lattice.

the_eff: the_eff (unitless)

Topological Hall effect efficiency.

thermK: K (W/mK)
Thermal conductivity.

ts_eff: ts (unitless)

Spin accumulation torque efficiency in the bulk.

tsi_eff: tsi (unitless)

Spin accumulation torque efficiency at interfaces.

Ym: Y (Pa)
Young’s modulus.

202
Commands – Essential

Essential commands only. These commands are used most often, or unlock a large
part of the functionality through interactive objects console output. For descriptions
see Commands – All section or type them in the console.

addconductor modules

addinsulator multiconvolution

addmesh ode

center params

chdir pbc

computefields reset

cuda run

data savesim

default setangle

display setfield

loadsim stages

mesh stop

203
Commands – Important

Important commands for more advanced users, but might not be used as often as
the essential commands. For descriptions see Commands – All section or type them
in the console.

ambient savemeshimage

curietemperature setcurrent

dwall setdefaultelectrodes

electrodes setdt

dp_getprofile setheatdt

iterupdate setpotential

paramstemp skyrmion

paramsvar showdata

preparemovingmesh temperature

resetmesh tsolverconfig

204
Commands – Useful

Other useful commands. For descriptions see Commands – All section or type them
in the console.

adddipole loadmaskfile

addelectrode makevideo

clearelectrodes refineroughness

clearroughness roughenmesh

copymeshdata scalemeshrects

copyparams setparamtemparray

dp_averagemeshrect surfroughenjagged

205
Commands – All (Alphabetical)

2dmulticonvolution

USAGE : 2dmulticonvolution status

Switch to multi-layered convolution and force it to 2D layering in each mesh (2), or 2D convolution for
each mesh (1), or allow 3D (0).

addafmesh

USAGE : addafmesh name rectangle

Add antiferromagnetic mesh with given name and rectangle (m). The rectangle can be specified as:
sx sy sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex ey ez with the start
point as the origin.

addameshcubic

USAGE : addameshcubic name rectangle

Add an atomistic mesh with simple cubic structure, with given name and rectangle (m). The rectangle
can be specified as: sx sy sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex
ey ez with the start point as the origin.

addconductor

USAGE : addconductor name rectangle

Add a normal metal mesh with given name and rectangle (m). The rectangle can be specified as: sx
sy sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex ey ez with the start point
as the origin.

adddata

USAGE : adddata dataname (meshname, (rectangle))

Add dataname to list of output data. If applicable specify meshname and rectangle (m) in mesh. If not
specified and required, active mesh is used with entire mesh rectangle.

206
adddiamagnet

USAGE : adddiamagnet name rectangle

Add a diamagnetic mesh with given name and rectangle (m). The rectangle can be specified as: sx sy
sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex ey ez with the start point as
the origin.

adddipole

USAGE : adddipole name rectangle

Add a rectangular dipole with given name and rectangle (m). The rectangle can be specified as: sx sy
sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex ey ez with the start point as
the origin.

addelectrode

USAGE : addelectrode electrode_rect

Add an electrode in given rectangle (m).

addinsulator

USAGE : addinsulator name rectangle

Add an insulator mesh with given name and rectangle (m). The rectangle can be specified as: sx sy
sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex ey ez with the start point as
the origin.

addmaterial

USAGE : addmaterial name rectangle

Add a new mesh with material parameters loaded from the materials database. The name is the
material name as found in the mdb file (see materialsdatabase command); this also determines the
type of mesh to create, as well as the created mesh name. The rectangle (m) can be specified as: sx
sy sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex ey ez with the start point
as the origin.
Script return values: meshname - return name of mesh just added (can differ from the material name).

207
addmdbentry

USAGE : addmdbentry meshname (materialname)

Add new entry in the local materials database from parameters in the given mesh. The name of the
new entry is set to materialname if specified, else set to meshname. For a complete entry you should
then edit the mdb file manually with all the appropriate fields shown there.

addmesh

USAGE : addmesh name rectangle

Add a ferromagnetic mesh with given name and rectangle (m). The rectangle can be specified as: sx
sy sz ex ey ez for the start and end points in Cartesian coordinates, or as: ex ey ez with the start point
as the origin.

addmodule

USAGE : addmodule meshname handle

Add module with given handle to named mesh.

addpinneddata

USAGE : addpinneddata dataname (meshname, (rectangle))

Add new entry in data box (at the end) with given dataname and meshname if applicable. A rectangle
may also be specified if applicable, however this will not be shown in the data box.

addrect

USAGE : addrect rectangle (meshname)

Fill rectangle (m) within given mesh (active mesh if name not given). The rectangle coordinates are
relative to specified mesh.

208
addstage

USAGE : addstage stagetype (meshname)

Add a generic stage type to the simulation schedule with name stagetype, specifying a meshname if
needed (if not specified and required, active mesh is used).

ambient

USAGE : ambient ambient_temperature (meshname)

Set mesh ambient temperature (all meshes if meshname not given) for Robin boundary conditions :
flux normal = alpha * (T_boundary - T_ambient).
Script return values: ambient_temperature - ambient temperature for mesh in focus.

astepctrl

USAGE : astepctrl err_fail err_high err_low dT_incr dT_min dT_max

Set parameters for adaptive time step control: err_fail - repeat step above this, err_high - decrease dT
abnove this, err_low - increase dT below this, dT_incr - increase dT using fixed multiplier, dT_min,
dT_max - dT bounds.

atomicmoment

USAGE : atomicmoment ub_multiple (meshname)

Set atomic moment as a multiple of Bohr magnetons (all applicable meshes if meshname not given)
for given mesh. This affects the temperature dependence of 'me' (see curietemperature command). A
non-zero value will result in me(T) being dependent on the applied field.
Script return values: ub_multiple - atomic moment multiple of Bohr magneton for mesh in focus.

averagemeshrect

USAGE : averagemeshrect (rectangle)

Calculate the average value depending on currently displayed quantities. The rectangle is specified in
relative coordinates to the currently focused mesh; if not specified average the entire focused mesh.
Script return values: value

209
benchtime

USAGE : benchtime

Show the last simulation duration time in ms, between start and stop; used for performance
becnhmarking.
Script return values: value

blochpreparemovingmesh

USAGE : blochpreparemovingmesh (meshname)

Setup the named mesh (or active mesh) for moving Bloch domain wall simulations: 1) set
movingmesh trigger, 2) set domain wall structure, 3) set dipoles left and right to remove end magnetic
charges, 4) enable strayfield module.

cellsize

USAGE : cellsize value

Change cellsize of mesh in focus (m). The cellsize can be specified as: hx hy hz, or as: hxyz
Script return values: cellsize - return cellsize of mesh in focus.

center

USAGE : center

Center mesh view and scale to fit window size.

chdir

USAGE : chdir directory

Change working directory.

Script return values: directory

210
checkupdates

USAGE : checkupdates

Connect to boris-spintronics.uk to check if updates to program or materials database are available.

clearelectrodes

USAGE : clearelectrodes

Delete all currently set electrodes.

clearequationconstants

USAGE : clearequationconstants

Clear all user-defined constants for text equations.

clearmovingmesh

USAGE : clearmovingmesh

Clear moving mesh settings made by a prepare command.

clearparamstemp

USAGE : clearparamstemp (meshname, (paramname))

Clear material parameter temperature dependence in given mesh. If meshname not given clear
temperature dependences in all meshes for all parameters. If paramname not given clear all
parameters temperature dependences in named mesh.

clearparamsvar

USAGE : clearparamsvar (meshname)

Clear all material parameters spatial dependence in given mesh. If meshname not given clear spatial
dependence in all meshes.

211
clearparamvar

USAGE : clearparamvar meshname paramname

Clear parameter spatial dependence in given mesh.

clearroughness

USAGE : clearroughness (meshname)

Clear roughness by setting the fine shape same as the coarse M shape.

clearscreen

USAGE : clearscreen

Clear all console text.

computefields

USAGE : computefields

Run simulation from current state for a single iteration without advancing the simulation time.

copymeshdata

USAGE : copymeshdata meshname_from meshname_to (...)

Copy all primary mesh data (e.g. magnetisation values and shape) from first mesh to all other meshes
given - all meshes must be of same type.

copyparams

USAGE : copyparams meshname_from meshname_to (...)

Copy all mesh parameters from first mesh to all other meshes given - all meshes must be of same
type.

212
coupletodipoles

USAGE : coupletodipoles status

Set/unset coupling to dipoles : if ferromagnetic meshes touch a dipole mesh then interface magnetic
cells are exchange coupled to the dipole magnetisation direction.

cuda

USAGE : cuda status

Switch CUDA GPU computations on/off.

Script return values: status

curietemperature

USAGE : curietemperature curie_temperature (meshname)

Set Curie temperature (all ferromagnetic meshes if meshname not given) for ferromagnetic mesh.
This will set default temperature dependencies as: Ms = Ms0*me, A = Ah*me^2, D = D0*me^2, K =
K0*me^3 (K1 and K2), damping = damping0*(1-T/3Tc) T < Tc, damping = damping0*2T/3Tc T >= Tc,
susrel = dme/d(mu0Hext). Setting the Curie temperature to zero will disable temperature dependence
for these parameters.
Script return values: curie_temperature - Curie temperature for mesh in focus.

data

USAGE : data

Shows list of currently set output data and available data.

Script return values: number of set output data fields

default

USAGE : default

Reset program to default state.

213
deldata

USAGE : deldata index

Delete data from list of output data at index number. If index number is -1 then delete all data fields,
leaving just a default time data field - there must always be at least 1 output data field.

delelectrode

USAGE : delelectrode index

Delete electrode with given index.

delequationconstant

USAGE : delequationconstant name

Delete named user constant used in text equations.

delmdbentry

USAGE : delmdbentry materialname

Delete entry in the local materials database (see materialsdatabase for current selection).

delmesh

USAGE : delmesh name

Delete mesh with given name.

delmodule

USAGE : delmodule meshname handle

Delete module with given handle from named mesh.

214
delpinneddata

USAGE : delpinneddata index

Delete entry in data box at given index (index in order of appearance in data box from 0 up).

delrect

USAGE : delrect rectangle (meshname)

Void rectangle (m) within given mesh (active mesh if name not given). The rectangle coordinates are
relative to specified mesh.

delstage

USAGE : delstage index

Delete stage from simulation schedule at index number. If index number is -1 then delete all stages,
leaving just a default Relax stage - there must always be at least 1 stage set.

designateground

USAGE : designateground electrode_index

Change ground designation for electrode with given index.

display

USAGE : display name (meshname)

Change quantity to display for given mesh (active mesh if name not given).

displaybackground

USAGE : displaybackground name (meshname)

Change background quantity to display for given mesh (active mesh if name not given).

215
displaythresholds

USAGE : displaythresholds minimum maximum

Set thresholds for foreground mesh display : magnitude values outside this range are not rendered. If
both set to 0 then thresholds are ignored.

displaythresholdtrigger

USAGE : displaythresholdtrigger trigtype

For vector quantities, set component to trigger thresholds on. trigtype = 1 (x component), trigtype = 2
(y component), trigtype = 3 (z component), trigtype = 5 (magnitude only)

displaytransparency

USAGE : displaytransparency foreground background

Set alpha transparency for display. Values range from 0 (fully transparent) to 1 (opaque). This is
applicable in dual display mode when we have a background and foreground for the same mesh.

dmcellsize

USAGE : dmcellsize value

Change demagnetizing field macrocell size of mesh in focus, for atomistic meshes (m). The cellsize
can be specified as: hx hy hz, or as: hxyz
Script return values: cellsize - return demagnetizing field macrocell size.

dp_add

USAGE : dp_add dp_source value (dp_dest)

Add value to dp array and place it in destination (or at same position if destination not specified).

dp_adddp

USAGE : dp_adddp dp_x1 dp_x2 dp_dest

Add dp arrays : dp_dest = dp_x1 + dp_x2

216
dp_append

USAGE : dp_append dp_original dp_new

Append data from dp_new to the end of dp_original.

dp_calcexchange

USAGE : dp_calcexchange

Calculate spatial dependence of exchange energy density for the focused mesh (must be magnetic
and have an exchange module enabled). Output available in Cust_S.

dp_calcsot

USAGE : dp_calcsot hm_mesh fm_mesh

For the given heavy metal and ferromagnetic meshes calculate the expected effective spin Hall angle
and field-like torque coefficient according to analytical equations (see manual).
Script return values: SHAeff, flST.

dp_calctopochargedensity

USAGE : dp_calctopochargedensity

Calculate topological charge density spatial dependence for the focused mesh (must be magnetic).
Output available in Cust_S.

dp_cartesiantopolar

USAGE : dp_cartesiantopolar dp_in_x dp_in_y (dp_out_r dp_out_theta)

Convert from Cartesian coordinates (x,y) to polar (r, theta).

dp_clear

USAGE : dp_clear indexes...

Clear dp arrays with specified indexes.

217
dp_clearall

USAGE : dp_clearall

Clear all dp arrays.

dp_coercivity

USAGE : dp_coercivity dp_index_x dp_index_y

Obtain coercivity from x-y data: find first crossings of x axis in the two possible directions, with
uncertainty obtained from step size.
Script return values: Hc_up Hc_up_err- Hc_up_err+ Hc_dn Hc_dn_err- Hc_dn_err+.

dp_completehysteresis

USAGE : dp_completehysteresis dp_index_x dp_index_y

For a hysteresis loop with only one branch continue it by constructing the other direction branch
(invert both x and y data and add it in continuation) - use only with hysteresis loops which are
expected to be symmetric.

dp_countskyrmions

USAGE : dp_countskyrmions (x y radius)

Calculate the number of skyrmions for focused mesh (must be magnetic), optionally in the given circle
with radius and centered at x y (relative values). Use Qmag = Integral(|m.(dm/dx x dm/dy)| dxdy) /
4PI.
Script return values: Q - the calculated topological charge.

dp_crossingsfrequency

USAGE : dp_crossingsfrequency dp_in_x dp_in_y dp_level dp_freq_up dp_freq_dn (steps)

From input x-y data build a histogram of average frequency the x-y data crosses a given line (up and
down, separated). The line varies between minimum and maximum of y data in given number of steps
(100 by default). Output the line values in dp_level with corresponding crossings frequencies in
dp_freq_up and dp_freq_dn.

218
dp_crossingshistogram

USAGE : dp_crossingshistogram dp_in_x dp_in_y dp_level dp_counts (steps)

From input x-y data build a histogram of number of times x-y data crosses a given line (up or down).
The line varies between minimum and maximum of y data in given number of steps (100 by default).
Output the line values in dp_level with corresponding number of crossings in dp_counts.

dp_div

USAGE : dp_div dp_source value (dp_dest)

Divide dp array by value and place it in destination (or at same position if destination not specified).
dp_divdp

USAGE : dp_divdp dp_x1 dp_x2 dp_dest

Divide dp arrays : dp_dest = dp_x1 / dp_x2

dp_dotprod

USAGE : dp_dotprod dp_vector ux uy uz dp_out

Take dot product of (ux, uy, uz) with vectors in dp arrays dp_vector, dp_vector + 1, dp_vector + 2 and
place result in dp_out.

dp_dotproddp

USAGE : dp_dotproddp dp_x1 dp_x2

Take dot product of dp arrays : value = dp_x1.dp_x2

dp_dumptdep

USAGE : dp_dumptdep meshname paramname max_temperature dp_index

Get temperature dependence of named parameter from named mesh up to max_temperature, at

dp_index - temperature scaling values obtained.

219
dp_erase

USAGE : dp_erase dp_index start_index length

From dp_index array erase a number of points - length - starting at start_index.

dp_extract

USAGE : dp_extract dp_in dp_out start_index (length)

From dp_in array extract a number of points - length - starting at start_index, and place them in
dp_out.

dp_fitadiabatic

USAGE : dp_fitadiabatic (abs_err Rsq T_ratio (stencil))

Fit the computed self-consistent spin torque (see below) using Zhang-Li STT with fitting parameters P
and beta (non-adiabaticity) using a given square in-plane stencil (default size 3) in order to extract the
spatial variation of P. Cut-off values for absolute fitting error (default 0.1), Rsq measure (default 0.9),
and normalized torque magnitude (default 0.1) can be set - value of zero disables cutoff. The focused
mesh must be ferromagnetic, have the transport module set with spin solver enabled, and we also
require Jc and either Ts or Tsi to have been computed. The fitting is done on Ts, Tsi, or on their sum
depending if they’ve been enabled or not. Output available in Cust_S.

dp_fitlorentz

USAGE : dp_fitlorentz dp_x dp_y

Fit Lorentz peak function to x y data : f(x) = y0 + S dH / (4(x-H0)^2 + dH^2).

Script return values: S, H0, dH, y0, std_S, std_H0, std_dH, std_y0.

dp_fitlorentz2

USAGE : dp_fitlorentz2 dp_x dp_y

Fit Lorentz peak function with both symmetric and asymmetric parts to x y data : f(x) = y0 + S (dH + A
* (x - H0)) / (4(x-H0)^2 + dH^2).
Script return values: S, A, H0, dH, y0, std_S, std_A, std_H0, std_dH, std_y0.

220
dp_fitnonadiabatic

USAGE : dp_fitnonadiabatic (abs_err Rsq T_ratio (stencil))

Fit the computed self-consistent spin torque (see below) using Zhang-Li STT with fitting parameters P
and beta (non-adiabaticity) using a given square in-plane stencil (default size 3) in order to extract the
spatial variation of beta. Cut-off values for absolute fitting error (default 0.1), Rsq measure (default
0.9), and normalized torque magnitude (default 0.1) can be set - value of zero disables cutoff. The
focused mesh must be ferromagnetic, have the transport module set with spin solver enabled, and we
also require Jc and either Ts or Tsi to have been computed. The fitting is done on Ts, Tsi, or on their
sum depending if they’ve been enabled or not. Output available in Cust_S.

dp_fitskyrmion

USAGE : dp_fitskyrmion dp_x dp_y

Fit skyrmion z component to obtain radius and center position : Mz(x) = Ms *

cos(2*arctan(sinh(R/w)/sinh((x-x0)/w))).
Script return values: R, x0, Ms, w, std_R, std_x0, std_Ms, std_w.

dp_fitsot

USAGE : dp_fitsot hm_mesh (rectangle)

Fit the computed self-consistent interfacial spin torque using SOT with fitting parameters SHAeff and
flST (field-like torque coefficient). hm_mesh specifies the heavy metal mesh from which to obtain the
current density. The fitting is done inside the specified rectangle for the focused mesh, with the
rectangle specified using relative coordinates as sx sy sz ex ey ez (entire mesh if not specified). The
focused mesh must be ferromagnetic, have the transport module set with spin solver enabled, and we
also require Jc and Tsi to have been computed.
Script return values: SHAeff, flST, std_SHAeff, std_flST, Rsq.

dp_fitsotstt

USAGE : dp_fitsotstt hm_mesh (rectangle)

Fit the computed self-consistent spin torque (see below) using Zhang-Li STT with fitting parameters P
and beta (non-adiabaticity), and simultaneously also using SOT with fitting parameters SHAeff and
flST (field-like torque coefficient). hm_mesh specifies the heavy metal mesh from which to obtain the
current density for SOT. The fitting is done inside the specified rectangle for the focused mesh, with

221
the rectangle specified using relative coordinates as sx sy sz ex ey ez (entire mesh if not specified).
The focused mesh must be ferromagnetic, have the transport module set with spin solver enabled,
and we also require Jc and either Ts or Tsi to have been computed to have been computed. The
fitting is done on Ts, Tsi, or on their sum depending if they’ve been enabled or not.
Script return values: SHAeff, flST, P, beta, std_SHAeff, std_flST, std_P, std_beta, Rsq.

dp_fitstt

USAGE : dp_fitstt (rectangle)

Fit the computed self-consistent spin torque (see below) using Zhang-Li STT with fitting parameters P
and beta (non-adiabaticity). The fitting is done inside the specified rectangle for the focused mesh,
with the rectangle specified using relative coordinates as sx sy sz ex ey ez (entire mesh if not
specified). The focused mesh must be ferromagnetic, have the transport module set with spin solver
enabled, and we also require Jc and either Ts or Tsi to have been computed. The fitting is done on
Ts, Tsi, or on their sum depending if they’ve been enabled or not.
Script return values: P, beta, std_P, std_beta, Rsq.

dp_get

USAGE : dp_get dp_arr index

Show value in dp_arr at given index - the index must be within the dp_arr size.
Script return values: value

dp_getampli

USAGE : dp_getampli dp_source pointsPeriod

Obtain maximum amplitude obtained every pointsPeriod points.

Script return values: amplitude.

dp_getexactprofile

USAGE : dp_getexactprofile start end step dp_index

Extract profile of physical quantity displayed on screen, directly from the mesh so using the exact
mesh resolution not the displayed resolution, along the line specified with given start and end
cartesian absolute coordinates (m), and with the given step size (m). If stencil specified - as x y z (m) -
then obtain profile values using weighted averaging with stencil centered on profile point. Place profile

222
in given dp arrays: 4 consecutive dp arrays are used, first for distance along line, the next 3 for
physical quantity so allow space for these starting at dp_index.

dp_getpath

USAGE : dp_getpath dp_index_in dp_index_out

Extract profile of physical quantity displayed on screen, directly from stored mesh data thus
independent of display resolution, along the path specified in Cartesian absolute coordinates (m)
through dp arrays at dp_index_in, dp_index_in + 1, dp_index_in + 2 (x, y, z coordinates resp.). Place
extracted profile in given dp arrays dp_index_out, dp_index_out + 1, dp_index_out + 2 (x, y, z
components for vector data).

dp_getprofile

USAGE : dp_getprofile start end dp_index

Extract profile of physical quantity displayed on screen, at the current display resolution, along the line
specified with given start and end cartesian absolute coordinates (m). Place profile in given dp arrays:
4 consecutive dp arrays are used, first for distance along line, the next 3 for physical quantity so allow
space for these starting at dp_index.

dp_histogram

USAGE : dp_histogram dp_x dp_y (bin min max)

Calculate a histogram with given bin, minimum and maximum values, from the magnetisation
magnitude of the focused mesh (must be magnetic). Save histogram in dp arrays at dp_x, dp_y. If
histogram parameters not given use a bin with 100 steps between minimum and maximum
magnetisation magnitude.

dp_histogram2

USAGE : dp_histogram2 dp_x dp_y (bin min max M2 deltaM2)

Calculate a histogram for a 2-sublattice mesh with given bin, minimum and maximum values for sub-
lattice A, if the corresponding magnetisation magnitude in sub-lattice B equals M2 within the given
deltaM2. Save histogram in dp arrays at dp_x, dp_y. If histogram parameters not given use a bin with
100 steps between minimum and maximum magnetisation magnitude, with M2 set to MeB and
deltaM2 set 0.01*MeB respectively.

223
dp_linreg

USAGE : dp_linreg dp_index_x dp_index_y (dp_index_z dp_index_out)

Fit using linear regression to obtain gradient and intercept with their uncertainties. If dp_index_z is
specified multiple linear regressions are performed on adjacent data points with same z value; output
in 5 dp arrays starting at dp_index_out as: z g g_err c c_err.
Script return values: g g_err c c_err.

dp_load

USAGE : dp_load (directory\)filename file_indexes... dp_indexes...

Load data columns from filename into dp arrays. file_indexes are the column indexes in filename (.txt
termination by default), dp_indexes are used for the dp arrays; count from 0. If directory not specified,
the default one is used.

dp_mean

USAGE : dp_mean dp_index

Obtain mean value with standard deviation.

Script return values: mean stdev.

dp_minmax

USAGE : dp_minmax dp_index

Obtain absolute minimum and maximum values, together with their index position.
Script return values: min_value min_index max_value max_index.

dp_monotonic

USAGE : dp_monotonic dp_in_x dp_in_y dp_out_x dp_out_y

From input x-y data extract monotonic sequence and place it in output x y arrays.

224
dp_mul

USAGE : dp_mul dp_source value (dp_dest)

Multiply dp array with value and place it in destination (or at same position if destination not specified).

dp_muldp

USAGE : dp_muldp dp_x1 dp_x2 dp_dest

Multiply dp arrays : dp_dest = dp_x1 * dp_x2

dp_newfile

USAGE : dp_newfile (directory/)filename

Make new file, erasing any existing file with given name. If directory not specified, the default one is
used.

dp_peaksfrequency

USAGE : dp_peaksfrequency dp_in_x dp_in_y dp_level dp_freq (steps)

From input x-y data build a histogram of average frequency of peaks in the x-y data in bands given by
the number of steps. The bands vary between minimum and maximum of y data in given number of
steps (100 by default). Output the line values in dp_level with corresponding peak frequencies in
dp_freq.
dp_rarefy

USAGE : dp_rarefy dp_in dp_out (skip)

Pick elements from dp_in using the skip value (1 by default) and set them in dp_out; e.g. with skip = 2
every 3rd data point is picked. The default skip = 1 picks every other point.

dp_remanence

USAGE : dp_remanence dp_index_x dp_index_y

Obtain remanence from x-y data: find values at zero field in the two possible directions.
Script return values: Mr_up Mr_dn.

225
dp_removeoffset

USAGE : dp_removeoffset dp_index (dp_index_out)

Subtract the first point (the offset) from all the points in dp_index. If dp_index_out not specified then
processed data overwrites dp_index.

dp_replacerepeats

USAGE : dp_replacerepeats dp_index (dp_index_out)

Replace repeated points from data in dp_index using linear interpolation: if two adjacent sets of
repeated points found, replace repeats between the mid-points of the sets. If dp_index_out not
specified then processed data overwrites dp_index.

dp_save

USAGE : dp_save (directory\)filename dp_indexes...

Save specified dp arrays in filename (.txt termination by default). If directory not specified, the default
one is used. dp_indexes are used for the dp arrays; count from 0.

dp_saveappend

USAGE : dp_saveappend (directory\)filename dp_indexes...

Save specified dp arrays in filename (.txt termination by default) by appending at the end. If directory
not specified, the default one is used. dp_indexes are used for the dp arrays; count from 0.

dp_saveappendasrow

USAGE : dp_saveappendasrow (directory\)filename dp_index

Save specified dp array in filename (.txt termination by default) as a single row with tab-spaced
values, appending to end of file. If directory not specified, the default one is used.

226
dp_saveasrow

USAGE : dp_saveasrow (directory\)filename dp_index

Save specified dp array in filename (.txt termination by default) as a single row with tab-spaced
values, appending to end of file. If directory not specified, the default one is used.

dp_sequence

USAGE : dp_sequence dp_index start_value increment points

Generate a sequence of data points in dp_index from start_value using increment.

dp_set

USAGE : dp_set dp_arr index value

Set value in dp_arr at given index - the index must be within the dp_arr size.

dp_showsizes

USAGE : dp_showsizes (dp_arr)

List sizes of all non-empty dp arrays, unless a specific dp_arr index is specified, in which case only
show the size of dp_arr.
Script return values: dp_arr size if specified

dp_smooth

USAGE : dp_smooth dp_in dp_out window_size

Smooth data in dp_in using nearest-neighbor averaging with given window size, and place result in
dp_out (must be different).
dp_sub

USAGE : dp_sub dp_source value (dp_dest)

Subtract value from dp array and place it in destination (or at same position if destination not
specified).

227
dp_subdp

USAGE : dp_subdp dp_x1 dp_x2 dp_dest

Subtract dp arrays : dp_dest = dp_x1 - dp_x2

dp_topocharge

USAGE : dp_topocharge (x y radius)

Calculate the topological charge for focused mesh (must be magnetic), optionally in the given circle
with radius and centered at x y (relative values). Q = Integral(m.(dm/dx x dm/dy) dxdy) / 4PI.
Script return values: Q - the calculated topological charge.

dwall

USAGE : dwall longitudinal transverse width position (meshname)

Create an idealised domain wall (tanh profile for longitudinal component, 1/cosh profile for transverse
component) along the x-axis direction in the given mesh (active mesh if name not specified). For
longitudinal and transverse specify the components of magnetisation as x, -x, y, -y, z, -z, i.e. specify
using these string literals. For width and position use metric units.

ecellsize

USAGE : ecellsize value

Change cellsize of mesh in focus for electrical conduction (m). The cellsize can be specified as: hx hy
hz, or as: hxyz
Script return values: cellsize - return electrical conduction cellsize of mesh in focus.

editdata

USAGE : editdata index dataname (meshname, (rectangle))

Edit entry in list of output data at given index in list. If applicable specify meshname and rectangle (m)
in mesh. If not specified and required, active mesh is used with entire mesh rectangle.

228
editdatasave

USAGE : editdatasave index savetype (savevalue)

Edit data saving condition in simulation schedule. Use index < 0 to set condition for all stages.

editstage

USAGE : editstage index stagetype (meshname)

Edit stage type from simulation schedule at index number.

editstagestop

USAGE : editstagestop index stoptype (stopvalue)

Edit stage/step stopping condition in simulation schedule. Use index < 0 to set condition for all stages.

editstagevalue

USAGE : editstagevalue index value

Edit stage setting value in simulation schedule. The value type depends on the stage type.

electrodes

USAGE : electrodes

Show currently configured electrodes.

equationconstants

USAGE : equationconstants name value

Create or edit user constant to be used in text equations.

229
errorlog

USAGE : errorlog status

Set error log status.

escellsize

USAGE : escellsize value

Change cellsize for electric super-mesh (m). The cellsize can be specified as: hx hy hz, or as: hxyz
Script return values: cellsize - return cellsize for electric super-mesh.

evalspeedup

USAGE : evalspeedup status

!!!Experimental!!!Status levels: 0 (no speedup), 1 (accurate), 2 (aggressive), 3 (extreme).

exchangecoupledmeshes

USAGE : exchangecoupledmeshes status (meshname)

Set/unset direct exchange coupling to neighboring meshes : if neighboring ferromagnetic meshes

touch the named mesh (set for focused mesh if meshname not given) then interface magnetic cells
are direct exchange coupled to them.
Script return values: status

excludemulticonvdemag

USAGE : excludemulticonvdemag status meshname

Set exclusion status (0 or 1) of named mesh from multi-layered demag convolution.

Script return values: status

flusherrorlog

USAGE : flusherrorlog

Clear error log.

230
fmscellsize

USAGE : fmscellsize value

Change cellsize for ferromagnetic super-mesh (m). The cellsize can be specified as: hx hy hz, or as:
hxyz
Script return values: cellsize - return cellsize for ferromagnetic super-mesh.

generate2dgrains

USAGE : generate2dgrains spacing (seed)

Generate 2D Voronoi cells in the xy plane at given average spacing. The seed is used for the pseudo-
random number generator, 1 by default.

generate3dgrains

USAGE : generate3dgrains spacing (seed)

Generate 3D Voronoi cells at given average spacing. The seed is used for the pseudo-random
number generator, 1 by default.

getvalue

USAGE : getvalue abspos

Get data value at abspos (absolute position in Cartesian coordinates) depending on currently
displayed quantities.
Script return values: value

imagecropping

USAGE : imagecropping left bottom right top

Set cropping of saved mesh images using normalized left, bottom, right, top values: 0, 0 point is left,
bottom of mesh window and 1, 1 is right, top of mesh window.

231
individualshape

USAGE : individualmaskshape status

When changing the shape inside a mesh, e.g. through a mask file, set this flag to true so the shape is
applied only to the primary displayed physical quantity. If set to false then all relevant physical
quantities are shaped.
Script return values: status

insulatingside

USAGE : insulatingside side_literal status (meshname)

Set temperature insulation (Neumann boundary condition) for named mesh side (active mesh if not
given). side_literal : x, -x, y, -y, z, -z.
Script return values: status_x status_-x status_y status_-y status_z status_-z - insulating sides status
for mesh in focus.

invertmag

USAGE : invertmag (components) (meshname)

Invert magnetisation direction. If mesh name not specified, the active mesh is used. You can choose
to invert just one or two components instead of the entire vector: specify components as x y z, e.g.
invertmag x

isrunning

USAGE : isrunning

Checks if the simulation is running and sends state value to the calling script.
Script return values: state - return simulation running state.

iterupdate

USAGE : iterupdate iterations

Update mesh display every given number of iterations during a simulation.

Script return values: iterations - return number of iterations for display update.

232
linkdtspeedup

USAGE : linkdtspeedup flag

Links speedup time-step to ODE time-step if set, else speedup time-step is independently controlled.
Applicable in extreme mode only.

linkdtstochastic

USAGE : linkdtstochastic flag

Links stochastic time-step to ODE time-step if set, else stochastic time-step is independently
controlled.

linkstochastic

USAGE : linkstochastic flag (meshname)

Links stochastic cellsize to magnetic cellsize if flag set to 1 for given mesh, else stochastic cellsize is
independently controlled. If meshname not given set for all meshes.

loadmaskfile

USAGE : loadmaskfile (z_depth) (directory\)filename

Apply .png mask file to magnetization in active mesh (i.e. transfer shape from .png file to mesh - white
means empty cells). If image is in grayscale then void cells up to given depth top down (z_depth > 0)
or down up (z_depth < 0). If z-depth = 0 then void top down up to all z cells.

loadovf2disp

USAGE : loadovf2disp (directory\)filename

Load an OOMMF-style OVF 2.0 file containing mechanical displacement data, into the currently
focused mesh (which must be ferromagnetic and have the melastic module enabled), mapping the
data to the current mesh dimensions. From the mechanical displacement the strain tensor is
calculated.

233
loadovf2mag

USAGE : loadovf2mag (renormalize_value) (directory\)filename

Load an OOMMF-style OVF 2.0 file containing magnetisation data, into the currently focused mesh
(which must be ferromagnetic), mapping the data to the current mesh dimensions. By default the
loaded data will not be renormalized: renormalize_value = 0. If a value is specified for
renormalize_value, the loaded data will be renormalized to it (e.g. this would be an Ms value).

loadovf2mesh

USAGE : loadovf2mesh (renormalize_value) (directory\)filename

Load an OOMMF-style OVF 2.0 file containing 3-component vector data. This will create a new
permalloy ferromagnetic mesh with dimensions and magnetization data obtained from the OVF 2.0
file. By default the loaded data will not be renormalized: renormalize_value = 0. If a value is specified
for renormalize_value, the loaded data will be renormalized to it (e.g. this would be an Ms value).

loadovf2strain

USAGE : loadovf2strain (directory\)filename_diag filename_odiag

Load an OOMMF-style OVF 2.0 file containing strain tensor data, into the currently focused mesh
(which must be ferromagnetic and have the melastic module enabled), mapping the data to the
current mesh dimensions. The symmetric strain tensor is applicable for a cubic crystal, and has 3
diagonal component (specified in filename_diag with vector data as xx, yy, zz), and 3 off-diagonal
components (specified in filename_odiag with vector data as yz, xz, xy).

loadsim

USAGE : loadsim (directory\)filename

Load simulation with given name.

makevideo

USAGE : makevideo (directory\)filebase fps quality

Make a video from .png files sharing the common filebase name. Make video at given fps and quality
(0 to 5 worst to best).

234
manual

USAGE : manual

Opens Boris manual for current version.

matcurietemperature

USAGE : matcurietemperature curie_temperature (meshname)

Set indicative material Curie temperature for ferromagnetic mesh (focused ferromagnetic mesh if
meshname not given). This is not used in calculations, but serves as an indicative value - set the
actual Tc value with the curietemperature command.
Script return values: curie_temperature - Indicative material Curie temperature for mesh in focus.

materialsdatabase

USAGE : materialsdatabase (mdbname)

Switch materials database in use. This setting is not saved by savesim, so using loadsim doesn't
affect this setting; default mdb set on program start.

mcellsize

USAGE : mcellsize value

Change cellsize of mesh in focus for mechanical solver (m). The cellsize can be specified as: hx hy
hz, or as: hxyz
Script return values: cellsize - return mechanical cellsize of mesh in focus.

memory

USAGE : memory

Show CPU and GPU-addressable memory information (total and free).

235
mesh

USAGE : mesh

Display information for all meshes.

meshfocus

USAGE : meshfocus meshname

Change mesh focus to given mesh name.

Script return values: meshname - return name of mesh in focus.

meshfocus2

USAGE : meshfocus2 meshname

Change mesh focus to given mesh name but do not change camera orientation.
Script return values: meshname - return name of mesh in focus.

meshrect

USAGE : meshrect rectangle

Change rectangle of mesh in focus (m). The rectangle can be specified as: sx sy sz ex ey ez for the
start and end points in Cartesian coordinates, or as: ex ey ez with the start point as the origin.
Script return values: rectangle - return rectangle of mesh in focus.

mirrormag

USAGE : mirrormag axis (meshname)

Mirror magnetisation in a given axis, specified as x, y, z, e.g. mirrormag x. If mesh name not specified,
the active mesh is used

modules

USAGE : modules

Show interactive list of available and currently set modules.

236
movingmesh

USAGE : movingmesh status_or_meshname

Set/unset trigger for movingmesh algorithm. If status_or_meshname = 0 then turn off, if

status_or_meshname = 1 then turn on with trigger set on first ferromagnetic mesh, else
status_or_meshname should specify the mesh name to use as trigger.

movingmeshasym

USAGE : movingmeshasym status

Change symmetry type for moving mesh algorithm: 1 for antisymmetric (domain walls), 0 for
symmetric (skyrmions).
Script return values: status

movingmeshthresh

USAGE : movingmeshthresh value

Set threshold used to trigger a mesh shift for moving mesh algorithm - normalised value between 0
and 1.
Script return values: threshold

multiconvolution

USAGE : multiconvolution status

Switch between multi-layered convolution (true) and supermesh convolution (false).

ncommon

USAGE : ncommon sizes

Switch to multi-layered convolution and force it to user-defined discretisation, specifying sizes as nx

ny nz.

237
ncommonstatus

USAGE : ncommonstatus status

Switch to multi-layered convolution and force it to user-defined discretisation (status = true), or default
discretisation (status = false).

neelpreparemovingmesh

USAGE : neelpreparemovingmesh (meshname)

Setup the named mesh (or active mesh) for moving Neel domain wall simulations: 1) set movingmesh
trigger, 2) set domain wall structure, 3) set dipoles left and right to remove end magnetic charges, 4)
enable strayfield module.

ode

USAGE : ode

Show interactive list of available and currently set ODEs and evaluation methods.

params

USAGE : params (meshname)

List all material parameters. If meshname not given use the active mesh.

paramstemp

USAGE : paramstemp (meshname)

List all material parameters temperature dependence. If meshname not given use the active mesh.

paramsvar

USAGE : paramsvar (meshname)

List all material parameters spatial variation. If meshname not given use the active mesh.

238
pbc

USAGE : pbc meshname flag images

Set periodic boundary conditions for magnetization in given mesh (must be ferromagnetic). Flags
specify types of perodic boundary conditions: x, y, or z; images specify the number of mesh images to
use either side for the given direction when calculating the demagnetising kernel - a value of zero
disables pbc. e.g. pbc x 10 sets x periodic boundary conditions with 10 images either side for the
focused mesh; pbc x 0 clears pbc for the x axis.

preparemovingmesh

USAGE : preparemovingmesh (meshname)

Setup the named mesh (or active mesh) for moving transverse (or vortex) domain wall simulations: 1)
set movingmesh trigger, 2) set domain wall structure, 3) set dipoles left and right to remove end
magnetic charges, 4) enable strayfield module.

random

USAGE : random (meshname)

Set random magnetisation distribution in mesh. If mesh name not specified, set for focused mesh.

refineroughness

USAGE : refineroughness value (meshname)

Set roughness refinement cellsize divider in given mesh, i.e. cellsize used for roughness initialization
is the ferromagnetic cellsize divided by value (3 components, so divide component by component).
Script return values: value - roughness refinement.

refreshmdb

USAGE : refreshmdb

Reload the local materials database (see materialsdatabase for current selection). This is useful if you
modify the values in the materials database file externally.

239
refreshscreen

USAGE : refreshscreen

Refreshes entire screen.

renamemesh

USAGE : renamemesh (old_name) new_name

Rename mesh. If old_name not specified then the mesh in focus is renamed.
requestmdbsync

USAGE : requestmdbsync materialname (email)

Request the given entry in the local materials database is added to the online shared materials
database. This must be a completed entry - see manual for instructions. The entry will be checked
before being made available to all users through the online materials database. If you want to receive
an update about the status of this request include an email address.

reset

USAGE : reset

Reset simulation state to the starting state.

resetmesh

USAGE : resetmesh (meshname)

Reset to constant magnetization in given mesh (active mesh if name not given).

robinalpha

USAGE : robinalpha robin_alpha (meshname)

Set alpha coefficient (all meshes if meshname not given) for Robin boundary conditions : flux normal
= alpha * (T_boundary - T_ambient).
Script return values: robin_alpha - Robin alpha value for mesh in focus.

240
roughenmesh

USAGE : roughenmesh depth (axis, (seed))

Roughen active mesh to given depth (m) along a named axis (use axis = x, y, or z as literal, z by
default). The seed is used for the pseudo-random number generator, 1 by default.

run

USAGE : run

Run simulation from current state.

savecomment

USAGE : savecomment (directory\)filename comment

Save comment in given file by appending to it.

savedatafile

USAGE : savedatafile (directory\)filename

Change output data file (and working directory if specified).

Script return values: filename

savedataflag

USAGE : savedataflag status

Set data saving flag status.

Script return values: status

saveimagefile

USAGE : saveimagefile (directory\)filename

Change image file base (and working directory if specified).

Script return values: filename

241
saveimageflag

USAGE : saveimageflag status

Set image saving flag status.

Script return values: status

savemeshimage

USAGE : savemeshimage ((directory\)filename)

Save currently displayed mesh image to given file (as .png). If directory not specified then default
directory is used. If filename not specified then default image save file name is used.

saveovf2

USAGE : saveovf2 (data_type) (directory\)filename

Save an OOMMF-style OVF 2.0 file containing data from the currently focused mesh. You can specify
the data type as data_type = bin4 (single precision 4 bytes per float), data_type = bin8 (double
precision 8 bytes per float), or data_type = text. By default bin8 is used.

saveovf2mag

USAGE : saveovf2mag (n) (data_type) (directory\)filename

Save an OOMMF-style OVF 2.0 file containing magnetisation data from the currently focused mesh
(which must be ferromagnetic). You can normalize the data to Ms0 value by specifying the n flag (e.g.
saveovf2mag n filename) - by default the data is not normalized. You can specify the data type as
data_type = bin4 (single precision 4 bytes per float), data_type = bin8 (double precision 8 bytes per
float), or data_type = text. By default bin8 is used.

saveovf2param

USAGE : saveovf2param (data_type) (meshname) paramname (directory\)filename

Save an OOMMF-style OVF 2.0 file containing the named parameter spatial variation data from the
named mesh (currently focused mesh if not specified). You can specify the data type as data_type =
bin4 (single precision 4 bytes per float), data_type = bin8 (double precision 8 bytes per float), or
data_type = text. By default bin8 is used.

242
savesim

USAGE : savesim (directory\)filename

Save simulation with given name. If no name given, the last saved/loaded file name will be used.

scalemeshrects

USAGE : scalemeshrects status

When changing a mesh rectangle scale and shift all other mesh rectangles in proportion if status set.
Script return values: status

scellsize

USAGE : scellsize value

Change cellsize of mesh in focus for stochastic properties (m). The cellsize can be specified as: hx hy
hz, or as: hxyz
Script return values: cellsize - return stochastic properties cellsize of mesh in focus.

scriptserver

USAGE : scriptserver status

Enable or disable the script communication server. When enabled the program will listen for
commands received using network sockets on port 1542.

setafmesh

USAGE : setafmesh name rectangle

Set a single antiferromagnetic mesh (deleting all other meshes) with given name and rectangle (m).
The rectangle can be specified as: sx sy sz ex ey ez for the start and end points in Cartesian
coordinates, or as: ex ey ez with the start point as the origin.

243
setameshcubic

USAGE : setameshcubic name rectangle

Set a single atomistic mesh (deleting all other meshes) with simple cubic structure, with given name
and rectangle (m). The rectangle can be specified as: sx sy sz ex ey ez for the start and end points in
Cartesian coordinates, or as: ex ey ez with the start point as the origin.

setangle

USAGE : setangle polar azimuthal (meshname)

Set magnetisation angle in mesh uniformly using polar coordinates. If mesh name not specified, this is
set for all ferromagnetic meshes.

setatomode

USAGE : setatomode equation evaluation

Set differential equation to solve in atomistic meshes, and method used to solve it (same method is
applied to micromagnetic and atomistic meshes).

setcurrent

USAGE : setcurrent current

Set a constant current source with given value. The potential will be adjusted to keep this constant
current.
Script return values: current

setdata

USAGE : setdata dataname (meshname, (rectangle))

Delete all currently set output data and set dataname to list of output data. If applicable specify
meshname and rectangle (m) in mesh. If not specified and required, active mesh is used with entire
mesh rectangle.

244
setdefaultelectrodes

USAGE : setdefaultelectrodes

Set electrodes at the x-axis ends of the given mesh, both set at 0V. Set the left-side electrode as the
ground. Delete all other electrodes.

setdisplayedparamsvar

USAGE : setdisplayedparamsvar meshname paramname

Set param to display for given mesh when ParamVar display is enabled (to show spatial variation if
any).

setdt

USAGE : setdt value

Set differential equation time-step (only applicable to fixed time-step methods).

Script return values: dT

setdtspeedup

USAGE : setdtspeedup value

Set time step for evaluation speedup, to be used in when in extreme mode.
Script return values: dTspeedup

setdtstoch

USAGE : setdtstoch value

Set time step for stochastic field generation.

Script return values: dTstoch

245
setelectrodepotential

USAGE : setelectrodepotential electrode_index potential

Set potential on electrode with given index.

Script return values: potential

setelectroderect

USAGE : setelectroderect electrode_index electrode_rect

Edit rectangle (m) for electrode with given index.

setfield

USAGE : setfield magnitude polar azimuthal (meshname)

Set uniform magnetic field (A/m) using polar coordinates. If mesh name not specified, this is set for all
magnetic meshes - must have Zeeman module added.
Script return values: <Ha_x, Ha_y, Ha_z> - applied field in Cartesian coordinates for mesh in focus.

setheatdt

USAGE : setheatdt value

Set heat equation solver time step.

Script return values: value - heat equation time step.

setmaterial

USAGE : setmaterial name

Set a single mesh with material parameters loaded from the materials database (deleting all other
meshes). The name is the material name as found in the mdb file (see materialsdatabase command);
this also determines the type of mesh to create, as well as the created mesh name. The rectangle (m)
can be specified as: <i>sx sy sz ex ey ez</i> for the start and end points in Cartesian coordinates, or
as: <i>ex ey ez</i> with the start point as the origin.
Script return values: Script return values: meshname - return name of mesh just added (can differ
from the material name).

246
setmesh

USAGE : setmesh name rectangle

Set a single ferromagnetic mesh (deleting all other meshes) with given name and rectangle (m). The
rectangle can be specified as: sx sy sz ex ey ez for the start and end points in Cartesian coordinates,
or as: ex ey ez with the start point as the origin.

setode

USAGE : setode equation evaluation

Set differential equation to solve in both micromagnetic and atomistic meshes, and method used to
solve it (same method is applied to micromagnetic and atomistic meshes).

setodeeval

USAGE : setodeeval evaluation

Set differential equation method used to solve it (same method is applied to micromagnetic and
atomistic meshes).

setparam

USAGE : setparam meshname paramname (value)

Set the named parameter to given value.

Script return values: value - return value of named parameter in named mesh.

setparamtemparray

USAGE : setparamtemparray meshname paramname filename

Set the named parameter temperature dependence using an array in the given mesh. This must
contain temperature values and scaling coefficients. Load directly from a file (tab spaced).

247
setparamtempequation

USAGE : setparamtempequation meshname paramname text_equation

Set the named parameter temperature dependence equation for the named mesh.

setparamvar

USAGE : setparamvar meshname paramname generatorname (arguments...)

Set the named parameter spatial dependence for the named mesh using the given generator
(including any required arguments for the generator - if not given, default values are used).

setpotential

USAGE : setpotential potential

Set a symmetric potential drop : -potential/2 for ground electrode, +potential/2 on all other electrodes.
Script return values: potential

setrect

USAGE : setrect polar azimuthal rectangle (meshname)

Set magnetisation angle in given rectangle of mesh (relative coordinates) uniformly using polar
coordinates. If mesh name not specified, the active mesh is used.
setsordamping

USAGE : setsordamping damping_v damping_s

Set fixed damping values for SOR algorithm used to solve the Poisson equation for V (electrical
potential) and S (spin accumulation) respectively.
Script return values: damping_v damping_s

setstage

USAGE : setstage stagetype (meshname)

Delete all currently set stages, and set a new generic stage type to the simulation schedule with name
stagetype, specifying a meshname if needed (if not specified and required, active mesh is used).

248
setstress

USAGE : setstress magnitude polar azimuthal (meshname)

Set uniform mechanical stress (Pa) using polar coordinates. If mesh name not specified, this is set for
all magnetic meshes - must have MElastic module added.
Script return values: <Tsig_x, Tsig_y, Tsig_z> - applied mechanical stress in Cartesian coordinates
for mesh in focus.

showa

USAGE : showa

Show predicted exchange stiffness (J/m) value for current mesh in focus (must be atomistic), using
formula A = J*n/2a, where n is the number of atomic moments per unit cell, and a is the atomic cell
size.
Script return values: A

showdata

USAGE : showdata dataname (meshname, (rectangle))

Show value(s) for dataname. If applicable specify meshname and rectangle (m) in mesh. If not
specified and required, active mesh is used with entire mesh rectangle.
Script return values: varies
showk

USAGE : showk

Show predicted uniaxial anisotropy (J/m^3) constant value for current mesh in focus (must be
atomistic), using formula K = k*n/a^3, where n is the number of atomic moments per unit cell, and a is
the atomic cell size.
Script return values: A

249
showlengths

USAGE : showlengths

Calculate a number of critical lengths for the focused mesh (must be ferromagnetic) to inform
magnetisation cellsize selection. lex = sqrt(2 A / mu0 Ms^2) : exchange length, l_Bloch = sqrt(A / K1) :
Bloch wall width, l_sky = PI D / 4 K1 : Neel skyrmion wall width.

showmcells

USAGE : showmcells

Show number of discretisation cells for magnetisation for focused mesh (must be ferromagnetic).
Script return values: n

showms

USAGE : showms

Show predicted saturation magnetisation (A/m) value for current mesh in focus (must be atomistic),
using formula Ms = mu_s*n/a^3, where n is the number of atomic moments per unit cell, and a is the
atomic cell size.
Script return values: Ms

showtc

USAGE : showtc

Show predicted Tc value (K) for current mesh in focus (must be atomistic), using formula Tc =
J*e*z/3kB, where e is the spin-wave correction factor, and z is the coordination number.
Script return values: Tc
skyrmion

USAGE : skyrmion core chirality diameter position (meshname)

Create an idealised Neel-type skyrmion with given diameter and centre position in the x-y plane (2
relative coordinates needed only) of the given mesh (active mesh if name not specified). Core
specifies the skyrmion core direction: -1 for down, 1 for up. Chirality specifies the radial direction
rotation: 1 for towards core, -1 away from core. For diameter and position use metric units.

250
skyrmionbloch

USAGE : skyrmionbloch core chirality diameter position (meshname)

Create an idealised Bloch-type skyrmion with given diameter and centre position in the x-y plane (2
relative coordinates needed only) of the given mesh (active mesh if name not specified). Core
specifies the skyrmion core direction: -1 for down, 1 for up. Chirality specifies the radial direction
rotation: 1 for clockwise, -1 for anti-clockwise. For diameter and position use metric units.

skyrmionpreparemovingmesh

USAGE : skyrmionpreparemovingmesh (meshname)

Setup the named mesh (or active mesh) for moving skyrmion simulations: 1) set movingmesh trigger,
2) set domain wall structure, 3) set dipoles left and right to remove end magnetic charges, 4) enable
strayfield module.

ssolverconfig

USAGE : ssolverconfig s_convergence_error (s_iters_timeout)

Set spin-transport solver convergence error and iterations for timeout (if given, else use default).
Script return values: s_convergence_error s_iters_timeout

stages

USAGE : stages

Shows list of currently set simulation stages and available stage types.
Script return values: number of set stages

startupscriptserver

USAGE : startupscriptserver status

Set startup script server flag.

251
startupupdatecheck

USAGE : startupupdatecheck status

Set startup update check flag.

statictransportsolver

USAGE : statictransportsolver status

If static transport solver is set, the transport solver is only iterated at the end of a stage or step. You
should set a high iterations timeout if using this mode.
Script return values: status

stochastic

USAGE : stochastic

Shows stochasticity settings : stochastic cellsize for each mesh and related settings.

stop

USAGE : stop

Stop simulation without resetting it.

surfroughenjagged

USAGE : surfroughenjagged depth spacing (seed, (sides))

Roughen active mesh surfaces using a jagged pattern to given depth (m) and peak spacing (m).
Roughen both sides by default, unless sides is specified as -z or z (string literal). The seed is used for
the pseudo-random number generator, 1 by default.

252
tau

USAGE : tau tau_11 tau_22 (tau_12 tau_21) (meshname)

Set ratio of exchange parameters to critical temperature (Neel) (all antiferromagnetic meshes if
meshname not given) for antiferromagnetic mesh. tau_11 and tau_22 are the intra-lattice
contributions, tau_12 and tau_21 are the inter-lattice contributions.
Script return values: tau_11 tau_22 tau_12 tau_21

tcellsize

USAGE : tcellsize value

Change cellsize of mesh in focus for thermal conduction (m). The cellsize can be specified as: hx hy
hz, or as: hxyz
Script return values: cellsize - return thermal conduction cellsize of mesh in focus.

temperature

USAGE : temperature value (meshname)

Set mesh base temperature (all meshes if meshname not given) and reset temperature. Also set
ambient temperature if Heat module added. If the base temperature setting has a spatial dependence
specified through cT, this command will take it into account but only if the Heat module is added. If
you want the temperature to remain fixed you can still have the Heat module enabled but disable the
heat equation by setting the heat dT to zero (setheatdt 0).
Script return values: value - temperature value for mesh in focus.

tmodel

USAGE : tmodel num_temperatures (meshname)

Set temperature model (determined by number of temperatures) in given meshname (focused mesh if
meshname not given). Note insulating meshes only allow a 1-temperature model.

253
tsolverconfig

USAGE : tsolverconfig convergence_error (iters_timeout)

Set transport solver convergence error and iterations for timeout (if given, else use default).
Script return values: convergence_error iters_timeout

updatemdb

USAGE : updatemdb

Switch to, and update the local materials database from the online shared materials database.

updatescreen

USAGE : updatescreen

Updates all displayed values on screen and also refreshes.

vecrep

USAGE : vecrep meshname vecreptype

Set representation type for vectorial quantities in named mesh (or supermesh). vecreptype = 0 (full),
vecreptype = 1 (x component), vecreptype = 2 (y component), vecreptype = 3 (z component),
vecreptype = 4 (direction only), vecreptype = 5 (magnitude only).

vortex

USAGE : vortex longitudinal rotation core (rectangle) (meshname)

Create a vortex domain wall with settings: longitudinal (-1: tail-to-tail, 1: head-to-head), rotation (-1:
clockwise, 1: counter-clockwise), core (-1: down, 1: up). The vortex may be set in the given rectangle
(entire mesh if not given), in the given mesh (focused mesh if not given).

254
Selected Publications using Boris

1. M.M. Vopson, M. Belusky, and S. Lepadatu “Diamagnetic coupling for

magnetic tuning in nano-thin films” Applied Physics Letters 116, 252402
(2020)

2. S. Lepadatu, “Efficient computation of demagnetizing fields for magnetic

multilayers using multi-layered convolution” Journal of Applied Physics 126,
103903 (2019)

3. M. Belusky, S. Lepadatu, J. Naylor, M.M. Vopson, “Study of roughness effect

in Fe and Co thin films prepared by plasma magnetron sputtering” Physica B
574, 411666 (2019)

4. S. Lepadatu, “Effect of inter-layer spin diffusion on skyrmion motion in

magnetic multilayers” Scientific Reports 9, 9592 (2019)

5. M. Belusky, S. Lepadatu, J. Naylor, M.M. Vopson, “Evidence of substrate

roughness surface induced magnetic anisotropy in Ni80Fe20 flexible thin
films” J. Magn. Magn. Mater. 478, 77 (2019)

6. S. Lepadatu, “Unified treatment of spin torques using a coupled magnetization

dynamics and three-dimensional spin current solver” Scientific Reports 7, 12937
(2017)

7. M.M. Vopson, J. Naylor, T. Saengow, E.G.Rogers, S. Lepadatu, Y.K. Fetisov,

"Development of flexible Ni80Fe20 magnetic nano-thin films" Physica B 525, 12
(2017)

8. S. Lepadatu, M.M. Vopson, “Heat assisted multiferroic solid-state memory”

Materials 10, 991 (2017)

9. S. Lepadatu, H. Saarikoski, R. Beacham, M.J.B. Romero, T.A. Moore, G. Burnell, S.

Sugimoto, D. Yesudas, M.C. Wheeler, J. Miguel, S.S. Dhesi, D. McGrouther, S.
McVitie, G. Tatara, and C.H. Marrows, “Very low critical current density for motion of

255
 coupled domain walls in synthetic ferrimagnet nanowires” Scientific Reports 7, 1640
(2017)

10. S. Lepadatu, “Interaction of Magnetization and Heat Dynamics for Pulsed Domain
Wall Movement with Joule Heating” Journal of Applied Physics 120, 163908 (2016)

11. S. Lepadatu, “Effective field model of roughness in magnetic nano-structures”

Journal of Applied Physics 118, 243908 (2015)

12. M.M. Vopson, S. Lepadatu, “Solving the electrical control of magnetic coercive field
paradox” Appl. Phys. Lett. 105, 122901 (2014)

256