abinit/doc/tutorial/nuc.md

306 lines
15 KiB
Markdown

---
authors: JWZ, XG
---
# Properties at the nucleus
## Observables near the atomic nucleus.
The purpose of this tutorial is to show how to compute several observables of
interest in Mössbauer, NMR, and NQR spectroscopy, namely:
* the electric field gradient,
* the isomer shift,
* chemical shielding
This tutorial should take about 2 hours.
[TUTORIAL_README]
## Electric field gradient
Various spectroscopies, including nuclear magnetic resonance and nuclear
quadrupole resonance (NMR and NQR), as well as Mössbauer spectroscopy, show
spectral features arising from the electric field gradient at the nuclear
sites. Note that the electric field gradient (EFG) considered here arises from
the distribution of charge within the solid, not due to any external electric fields.
The way that the EFG is observed in spectroscopic experiments is through its
coupling to the nuclear electric quadrupole moment. The physics of this
coupling is described in various texts, for example [[cite:Slichter1978]].
Abinit computes the field gradient at each site, and then reports the gradient and its
coupling based on input values of the nuclear quadrupole moments.
The electric field and its gradient at each nuclear site arises from the
distribution of charge, both electronic and ionic, in the solid. The gradient
especially is quite sensitive to the details of the distribution at short
range, and so it is necessary to use the PAW formalism to compute the gradient
accurately. For charge density $n({\mathbf r})$, the potential $V$ is given
by
$$ V({\mathbf r})=\int \frac{n({\mathbf r'})}{ |\mathbf{r}-\mathbf{r'}| } d{\mathbf r'} $$
and the electric field gradient is
$$ V_{ij} = -\frac{\partial^2}{\partial x_i\partial x_j}V({\mathbf r}). $$
The gradient is computed at each nuclear site, for each source of charge arising
from the PAW decomposition (see [the tutorial PAW1](/tutorial/paw1) ).
This is done in the code as follows [[cite:Profeta2003]],[[cite:Zwanziger2008]]:
* Valence space described by planewaves: expression for gradient is Fourier-transformed at each nuclear site.
* Ion cores: gradient is computed by an Ewald sum method
* On-site PAW contributions: moments of densities are integrated in real space around each atom, weighted by the gradient operator
The code reports each contribution separately if requested.
The electric field gradient computation is performed at the end of a ground-state calculation,
and takes almost no additional time.
The tutorial file is for stishovite, a polymorph of SiO$_2$. In addition to typical ground state
variables, only two additional variables are added:
nucefg 2
quadmom 0.0 -0.02558
{% dialog tests/tutorial/Input/tnuc_1.abi %}
The first variable instructs Abinit to compute and print the electric field
gradient, and the second gives the quadrupole moments of the nuclei, in
barns, one for each type of atom.
A standard source for quadrupole moments is [[cite:Pyykko2008]].
Here we are considering silicon and oxygen, and in
particular Si-29, which has zero quadrupole moment, and O-17, the only stable
isotope of oxygen with a non-zero quadrupole moment.
After running the file *tnuc_1.abi* through Abinit, you can find the following
near the end of the output file:
Electric Field Gradient Calculation
atom : 1 typat : 1
Nuclear quad. mom. (barns) : 1.0000 Cq (MHz) : 0.0000 eta : 0.0000
efg eigval (au) : -0.152323 ; (V/m^2) : -1.48017693E+21
- eigvec : -0.000000 -0.000000 -1.000000
efg eigval (au) : -0.054274 ; (V/m^2) : -5.27401886E+20
- eigvec : 0.707107 -0.707107 0.000000
efg eigval (au) : 0.206597 ; (V/m^2) : 2.00757882E+21
- eigvec : 0.707107 0.707107 -0.000000
total efg : 0.076161 0.130436 -0.000000
total efg : 0.130436 0.076161 -0.000000
total efg : -0.000000 -0.000000 -0.152323
efg_el : 0.095557 0.004024 -0.000000
efg_el : 0.004024 0.095557 -0.000000
efg_el : -0.000000 -0.000000 -0.191114
efg_ion : -0.099183 0.005966 0.000000
efg_ion : 0.005966 -0.099183 0.000000
efg_ion : 0.000000 0.000000 0.198365
efg_paw : 0.079787 0.120445 0.000000
efg_paw : 0.120445 0.079787 0.000000
efg_paw : 0.000000 0.000000 -0.159574
This fragment gives the gradient at the first atom, which was silicon. Note
that the gradient is not zero, but the coupling is---that's because the
quadrupole moment of Si-29 is zero, so although there's a gradient there's
nothing in the nucleus for it to couple to.
Atom 3 is an oxygen atom, and its entry in the output is:
atom : 3 typat : 2
Nuclear quad. mom. (barns) : -0.0256 Cq (MHz) : 6.6150 eta : 0.1403
efg eigval (au) : -1.100599 ; (V/m^2) : -1.06949233E+22
- eigvec : 0.707107 -0.707107 0.000000
efg eigval (au) : 0.473085 ; (V/m^2) : 4.59714112E+21
- eigvec : 0.000000 0.000000 -1.000000
efg eigval (au) : 0.627514 ; (V/m^2) : 6.09778216E+21
- eigvec : 0.707107 0.707107 0.000000
total efg : -0.236543 0.864057 0.000000
total efg : 0.864057 -0.236543 0.000000
total efg : 0.000000 0.000000 0.473085
efg_el : -0.036290 -0.075078 0.000000
efg_el : -0.075078 -0.036290 0.000000
efg_el : 0.000000 0.000000 0.072579
efg_ion : -0.016807 0.291185 -0.000000
efg_ion : 0.291185 -0.016807 -0.000000
efg_ion : -0.000000 -0.000000 0.033615
efg_paw : -0.183446 0.647950 0.000000
efg_paw : 0.647950 -0.183446 0.000000
efg_paw : 0.000000 0.000000 0.366891
Now we see the electric field gradient coupling, in frequency units, along
with the asymmetry of the coupling tensor, and, finally, the three
contributions to the total. Note that the valence part, efg_el, is
small, while the ionic part and the on-site PAW part are larger. In fact, the
PAW part is largest; this is why these calculations give very poor results
with norm-conserving pseudopotentials, and need the full accuracy of PAW to capture
the behavior near the nucleus.
Experimentally, the nuclear quadrupole coupling for O-17 in stishovite is
reported as $6.5\pm 0.1$ MHz, with asymmetry $0.125\pm 0.05$ [[cite:Xianyuxue1994]].
It is not uncommon for PAW-based EFG calculations to give coupling values a few percent
too large; often this can be improved by using PAW datasets with smaller PAW
radii, at the expense of more expensive calculations [[cite:Zwanziger2016]].
## Fermi contact interaction
The Fermi contact interaction arises from overlap of the electronic wavefunctions
with the atomic nucleus, and is an observable for example in
Mössbauer spectroscopy [[cite:Greenwood1971]]. In Mössbauer spectra,
the isomer shift $\delta$ is expressed in (SI) velocity units as
$$ \delta = \frac{2\pi}{3}\frac{c}{E_\gamma}\frac{Z e^2}{ 4\pi\epsilon_0} ( |\Psi (R)_A|^2 - |\Psi (R)_S|^2 )\Delta\langle r^2\rangle $$
where $\Psi(R)$ is the electronic
wavefunction at nuclear site $R$, for the absorber (A) and source (S) respectively;
$c$ is the speed of light, $E_\gamma$ is the nuclear transition energy, and $Z$ the atomic number;
and $\Delta\langle r^2\rangle$ the change in the nuclear size squared. All these quantities
are assumed known in the Mössbauer spectrum of interest, except $|\Psi(R)|^2$, the
Fermi contact term.
Abinit computes the Fermi contact term in the PAW formalism by using as observable
$\delta(R)$, that is, the Dirac delta function at the nuclear site [[cite:Zwanziger2009]].
Like the electric field gradient computation, the Fermi contact calculation
is performed at the end of a ground-
state calculation, and takes almost no time. There is a tutorial file for
SnO$_2$, which, like stishovite studied above, has the rutile structure.
In addition to typical ground state
variables, only one additional variable is needed:
nucfc 1
{% dialog tests/tutorial/Input/tnuc_2.abi %}
After running this file, inspect the output and look for the phrase
"Fermi-contact Term Calculation". There you'll find the FC output for
each atom; in this case, the Sn atoms, [[typat]] 1, yield a contact term
of 71.6428 (density in atomic units, $a^{-3}_0$).
To interpret Mössbauer spectra you need really both a source and
an absorber; in the tutorial we provide also a file for $\alpha$-Sn (grey
tin, which is non-metallic).
{% dialog tests/tutorial/Input/tnuc_3.abi %}
If you run this file, you should find a contact term of 102.0748.
To check your results, you can use experimental data for the isomer shift $\delta$
for known compounds to compute $\Delta\langle r^2\rangle$ in the above equation
(see [[cite:Zwanziger2009]]). Using our results above together with standard
tin Mössbauer parameters of $E_\gamma = 23.875$ keV and an experimental shift
of 2.2 mm/sec for $\alpha$-Sn relative to SnO$_2$, we find
$\Delta\langle r^2\rangle = 5.67\times 10^{-3}\mathrm{fm}^2$, in decent agreement
with other calculations of 6--7$\times 10^{-3}\mathrm{fm}^2$ [[cite:Svane1987]], [[cite:Svane1997]].
## Chemical shielding
The chemical shielding is routinely measured in NMR spectroscopy, and arises from the orbital
currents induced by an external magnetic field. From an energetic point of view, it can be
understood as the joint response between an external magnetic field, and the nuclear magnetic
dipole moment at an atomic site, that is,
$$ \sigma_{ij} = \frac{\partial^2 E}{\partial m_i\partial B_j} $$
for shielding $\sigma$ and magnetic dipole $m$.
The total energy could then be viewed as
$$ E = E^0 - m\cdot B + m_i\sigma_{ij}B_j, $$
where the first term is the unperturbed energy, the second is the Zeeman interaction,
and the third involves the shielding.
The effect can thus be viewed either as a bare dipole interacting with a
shielded magnetic field (the traditional view),
or conversely [[cite:Thonhauser2009]] [[cite:Ceresoli2010]],
a shielded dipole interacting with the bare magnetic field. In Abinit we adopt
the second view point, and so compute the perturbed energy $\partial E/\partial B$ in the
presence of a nuclear dipole moment. The necessary expressions are found in [[cite:Zwanziger2023]], see also [[cite:Gonze2011a]] [[cite:Ceresoli2006]], and
are quite complex; in highly abbreviated form we evaluate
$$ \partial E/\partial B_\alpha = -M_{\alpha} = \frac{i}{2}\epsilon_{\alpha\beta\gamma}\sum_n^{\mathrm{occ}} \int d\mathbf{k}
\langle \partial_{k_\beta} u_{n\mathbf{k}}|(H_{\mathbf{k}}+E_{\mathbf{k}})|\partial_{k_\gamma} u_{n\mathbf{k}}\rangle $$
for the induced magnetic moment $M$ in direction $\alpha$. Notice the implied summation over the antisymmetric unit tensor $\epsilon_{\alpha\beta\gamma}$;
the structure is hence that of a cross product, which yields the circulation induced by the Hamiltonian.
In the full PAW treatment implemented in Abinit there are a number of other related terms as well. Note
in this expression the appearance of the derivatives of the wavefunctions; these are obtained from the DDK
perturbation in Abinit, as was examined in
[tutorial DFPT1](/tutorial/rf1) ).
Carrying out the calculation in Abinit is a two-step process: first, the ground state wavefunctions must be calculated
in the presence of a nuclear magnetic dipole moment, on the atom of interest in the direction of interest; then these
wavefunctions are used to compute the DDK wavefunctions under the same conditions, followed by assembly of the terms
making up $\sigma_{ij}$. Here is a sample input file:
{% dialog tests/tutorial/Input/tnuc_4.abi %}
This file is for an isolated neon atom, that is, a neon atom at the
center of a large box. There are two chained calculations: first, the
ground state, and then, the DDK perturbation ([[rfelfd]] = 2 or
equivalently [[rfddk]] = 1). In both cases, a nuclear magnetic dipole
moment is applied to the neon atom with the input variable
[[nucdipmom]]. The orbital magnetic moment calculation is triggered at
the end of the DDK calculation with
orbmag2 2
Both calculations must be done with the nuclear dipole moment on the
atom of interest, in the direction of interest. Here, there is only a
single atom, and due to spherical symmetry, all directions are
equivalent, so it is sufficient to use
nucdipmom 1.0 0.0 0.0
Notice that we use a dipole moment of size 1; this is purely for
convenience as we will see below. Running this file should take 3
minutes or so. After completion you will find in the output file:
Orbital magnetic moment computed with DFPT derivative wavefunctions
Orbital magnetic moment, Cartesian directions :
-5.54877537E-04 6.30837909E-15 4.28342968E-15
Chern vector, Cartesian directions :
-2.52162344E-09 1.10115145E-17 6.96089655E-18
The first result is the induced magnetic moment, which is a
vector--note that it points in the original dipole direction. To get
the shielding, divide the dipole strength (here, 1), and multiply by
(-1) to get the normal NMR sign convention--the result is
554.9~ppm. This result is underconverged--the fully converged result,
with [[ecut]]=30, is 552.1 ppm. For comparison, the all-electron,
wavefunction-based calculation using post-Hartree-Fock CCSD(T) of
[[cite:Vaara2003relativistic]] yields 551.9 ppm, so clearly the
present method is capable of excellent accuracy. The second line is
the Chern vector [[cite:Ceresoli2006]], that is, the integral of the
Berry curvature, which outside of exotic topological insulators,
should be zero. We include it as a convergence check. In this fast
test calculation we obtained $-2.5\times 10^{-9}$; in fully converged
calculations one would like numbers about 3 orders of magnitude
smaller.
A final detail to note: the induced moment due to the filled core
orbitals, is due to what is called the Lamb shielding. Usual PAW
datasets do not include this value, but you can add it to your input
file with the variable [[lambsig]]. To compute it you must obtain the
core wavefunctions, which is a normal option in Atompaw (see [tutorial
PAW2](/tutorial/paw2)). Then, you can integrate the core wavefunctions
to generate the Lamb shielding, through
$$ \sigma_{Lamb} = \frac{\alpha^2}{3}\sum^{cores}_n \left\langle\psi_n|\frac{1}{r}|\psi_n\right\rangle $$
where $\alpha$ is the fine structure constant.
Moving forward, should you want to compute the shielding tensor for an
atomic site in a solid, the procedure is identical, except that you
will have to run the sequence three times, once for each direction of
the applied dipole. Then you will have obtained three component
vectors $M_\alpha$, $M_\beta$, and $M_\gamma$, which you can assemble
into a $3\times 3$ matrix, and diagonalize to find the principal
values and directions. The result is the shielding tensor at the site
of interest, and orientated with respect to the Cartesian directions.