Orbital magnetic susceptibility of multifold fermions.

Topological semimetals demonstrate a variety of interesting properties (see, e.g., the review articles [1, 2, 3, 4, 5]). In the most simple case of Weyl semimetal, electronic spectrum near the Fermi level posses nodes with linear crossing of two bands. The absolute value of topological charge (or Chern number) of a Weyl node is equal to unity. Low energy excitations around such nodes correspond to fermions with (pseudo-)spin 1/2 and can be described by the Hamiltonian

As was shown in [6], materials of certain crystal structures contain nodes with linear crossing of multiple bands. The low energy excitations around these nodes were named unconventional fermions [6]. The fermions described by Hamiltonian (1) with pseudospin $J>1/2$ belong to this group. One of examples of such materials is CoSi, crystallizing in cubic noncentrosymmetric structure B20 (space group $P2_{1}3$, No.198)[7, 8]. In CoSi, topological band crossings exist at the time-reversal invariant $\Gamma$ and $R$ points. Their topological charge is 4, and there are four elongated Fermi arcs connecting projections of these points onto the surface Brillouin zone. The peculiarities of their band structure were confirmed by ARPES measurements [9, 10, 11]. Without spin-orbit coupling the low-energy excitations near the $\Gamma$ point correspond to fermions with pseudospin $J=1$. Spin-orbit coupling splits the spectrum into a Weyl node and a node with 4-fold linear band crossing [7, 8]. If one considers spin-orbit coupling in the zeroth order in $\mathbf{k}$ around the $\Gamma$ point, the spectrum of these 4-fold degenerate states can be described by pseudospin-3/2 Hamiltonian. The list of space groups that can host symmetry protected fermions with the pseudospin-3/2 were given in [6]. Whether materials with $J\geq 2$ can exist is still unknown [12]. The analysis in [6] showed that there are space groups that can host up to eight-fold linear band crossing points but not all of them can be represented by the Hamiltonian of considered form (1). For example, it was noted that fermions with pseudospin $J\geq 2$ cannot exist at high symmetry points of the Brillouin zone [6].

The nontrivial topology of band structure leads to a number of interesting transport properties in magnetic field, such as chiral anomaly and negative magnetoresistance [13, 3]. They are associated with the spectrum of Landau levels, that contains the chiral energy level, where carriers can move only along or against magnetic field, depending on the sign of the topological charge. In cobalt monosilicide, these effects were studied in the work [14]. Quantum oscillations of resistance and thermopower were studied in CoSi in magnetic fields up to 15 T [15, 16, 17].

Cobalt monosilicide and isostructural materials exhibit a wide variety of magnetic properties. CoSi single crystals at temperatures above 25 K are diamagnetic, but at lower temperatures the magnetic susceptibility of the most clear samples changes sign [25, 26]. CoSi solid solutions with a small Fe content are diamagnetic at room temperature, however with increasing iron content, ferromagnetic properties appear [27]. In the isostructural material MnSi in low magnetic fields, a helical magnetic structure was revealed, which, with increasing field, turns into a ferromagnetic structure [28, 29]. In MnSi, the formation of a skyrmion phase is also possible [30].

The magnetic susceptibility of topological semimetals also has a number of interesting features. The susceptibility of Weyl semimetals was studied in a number of works reviewed in [18]. A giant diamagnetic anomaly of orbital susceptibility in Weyl semimetal was predicted in [19]. It is characterized by a logarithmic divergence of susceptibility as the chemical potential approaches the Weyl node energy. This anomaly in Weyl and Dirac semimetals was also studied in [20, 21]. For an arbitrary spectrum slope, it was analyzed in the work [22]. The temperature dependence of susceptibility turns out to be nonmonotonic and has a minimum at the temperature $T\sim 0.443\mu/k_{B}$ [18], where the chemical potential $\mu$ is measured from the node energy. Interestingly, the minimum of susceptibility was observed in the Weyl semimetal TaAs at approximately 185 K [23]. A study of magnetization in TaAs [24] also showed that, in contrast to quasiparticles with a nonrelativistic spectrum, there is no saturation of longitudinal magnetization in strong fields in a Weyl semimetal.

The orbital magnetic susceptibility is determined by peculiarities of Landau levels in magnetic field. For topological semimetals, in which a topological charge greater than unity is associated with a nonlinear dependence of energy on the wave vector in some directions, the spectrum in a magnetic field and the density of states were calculated in [31]. For fermions with pseudospin 1 and a linear dispersion law, the spectrum in a magnetic field was given in the appendix to the work [6]. For fermions with higher pseudospin values, Landau levels were calculated in [12]. However, magnetic susceptibility was not considered in these works.

Orbital magnetic susceptibility of fermions with pseudospin 1 were considered in [32] and applied to the case of CoSi without spin-orbital coupling. It was shown that the contribution of pseudospin-1 states demonstrate a transition from negative (diamagnetic) to large positive (paramagnetic) susceptibility when the chemical potential approaches the energy of the node due to the contribution of the flat band states.

In the present work we extend analysis of our previous paper [32] to the case of orbital magnetic susceptibility of fermions with pseudospin greater than unity, and we study the susceptibility of multifold fermions in CoSi near the $\Gamma$ point of the Brillouin zone with the account of spin-orbit coupling.

Let us consider the fermions described by the Hamiltonian (1), generalized to the case of $J\geq 1/2$. The spectrum of fermions in magnetic field can be calculated using Pierls substitution. We followed the approach of [12], slightly reformulated using matrix notation. The momentum operator $\hat{\mathbf{p}}$ is replaced with $\hat{\mathbf{\pi}}=\hat{\mathbf{p}}+e\mathbf{A}/c$, where $\mathbf{A}$ is the vector potential of magnetic field that is equal to $\mathbf{A}=(0,Bx,0)$ in Landau gauge. It is convenient to introduce the rising and lowering operators $\hat{J}_{\pm}=\hat{J}_{x}\pm i\hat{J}_{y}$ and $\hat{a}^{+}=(\hat{\pi}_{x}+i\hat{\pi}_{y})/(i\sqrt{2b}\hbar)$, $\hat{a}=-(\hat{\pi}_{x}-i\hat{\pi}_{y})/(i\sqrt{2b}\hbar)$. Here, $b=eB/\hbar c$ is proportional to magnetic field and is related to the magnetic length $l$ as $b=l^{-2}$. As a result, the Hamiltonian in magnetic field takes the form

The multicomponent wave function $\psi_{j}(x)$ ($j=J,J-1,...,-J$) can be taken as a superposition of eigenfunctions of harmonic oscillator $\phi_{n}(x)$ [33]

Multicomponent wave functions of the chiral levels have one or several zero components [12]. For Weyl fermions, it is $\psi_{1/2}$. In the case of pseudospin-1 fermions, there are two chiral levels and the zero components are either $\psi_{1}$ or $\psi_{1}$ and $\psi_{0}$. To account for such states in general case, it is convenient to extend low bound of summation in (3) down to $n=-(J-j)$, understanding that $c_{jn}=0$ for $n<0$. The substitution of (3) into the Schroedinger equation leads to the matrix eigenvalue problem. It can be reduced to closed systems of $2J+1$ equations for each $n$ that have the form $\hat{\mathcal{H}}\,\mathbf{c}=\epsilon\,\mathbf{c}$, where $\mathbf{c}=\left(c_{J,n},c_{J-1,n+1},...,c_{-J,n+2J}\right)^{\mathrm{T}}$. The matrix $\hat{\mathcal{H}}$ is tridiagonal with the matrix elements $\mathcal{H}_{j,j}=\hbar\,v\,k_{z}j$ and $\mathcal{H}_{j,j+1}=\mathcal{H}_{j+1,j}^{*}=i\,\hbar\,v\,(\hat{J}_{+})_{j,j+1}((n+J-j)\,b/2)^{1/2}$ for $j=J,J-1,...-J$, where $(\hat{J}_{+})_{j,j\pm 1}=(J(J+1)-j(j\pm 1))^{1/2}$. For each $n\geq 0$ one obtains $2J+1$ subbands of Landau levels $\epsilon_{m}(n,k_{z},b)$, $m=J,J-1,...,-J$. For $-2J\leq n<0$ first $|n|$ rows and columns of matrix $\hat{\mathcal{H}}$ should be discarded and additional $J(2J+1)$ midgap solutions appear.

As an example, the spectra of Landau levels for $J=1$ and $3/2$ are plotted in the figure 1 for $B=14$T. In estimations we used $v_{0}=18.6\cdot 10^{6}$cm/s and the maximum wave vector $k_{0}=0.08$Å^-1 that are typical for the states with $J=1$ near $\Gamma$ point in CoSi [8]. To make the range of energies for different $J$ comparable, we used electron velocities equal to $v=v_{0}/J$. The Landau levels for $n\geq 0$ are plotted with solid lines. The midgap solutions for $n<0$ is convinient to denote by a pair of indexes $(m,\,m^{\prime})$ of bands that they connect. These solutions are plotted with dashed lines. For $J=1$ there are three bands of Landau levels and three midgap solutions. For $J=3/2$ there are four bands of Landau levels and six midgap solutions.

Magnetic susceptibility $\chi$ is determined by all states filled with electrons, but its dependence on temperature and chemical potential is determined by electronic states with energy close to $\mu$. The influence of topological states will be important if the chemical potential is close to the band crossing energy. Following [19, 20, 21], we split the susceptibility into the contribution from the energy range $\pm\epsilon_{0}$ around the band crossing and a background contribution from the other parts of the electronic spectrum. The contribution of topological states to the low-field magnetic susceptibility of Weyl semimetals of the first and second types were considered in [19, 20, 21]. In [32], the case of $J=1$ was considered. Here we extend the results to the case of $J>1$.

Using the obtained electronic spectrum in magnetic field the thermodynamic potential $\Omega$ of the unit volume can be calculated. Then the orbital magnetic susceptibility is defined as $\chi=-\partial^{2}\Omega/\partial B^{2}=-(e/\hbar c)^{2}\partial^{2}\Omega/\partial b^{2}$.

Substituting the powers of $\delta$ with corresponding average values in (6), gives

Using this equation allows to evaluate orbital susceptibility analytically for the case of $J\leq 3/2$, when required derivatives can be calculated using the secular equation.

For $J=1$, there are three midgap solutions (fig. 1, left panel) that can be denoted by a pair of indexes $(m,\,m^{\prime})$ of the bands that they connect. The first midgap solution $(-1,1)$ does not depend on magnetic field, but others ($(1,0)$, $(0,-1)$) do. It is convenient to combine together the contributions of bands $m=\pm 1$, the midgap solution $(-1,1)$ and two halves of midgap solutions $(1,0)$ for $k_{z}<0$ and $(0,-1)$ for $k_{z}>0$. This contribution has the same form as Eq. (8) for the Weyl node [32], if one replaces $v$ by $2v$, so that the slopes of the bands were the same.

In the case of pseudospin $J=1$, there is also a contribution from dispersionless band ($m=0$) and midgap solutions $(1,0)$ for $k_{z}>0$ and $(0,-1)$ for $k_{z}<0$. They lie in the bounded energy interval that leads to high density of states. In this case it is not possible to determine $n_{m}$ from the limiting energy $\epsilon_{0}$. The number of levels in this group can be obtained from the conservation of the total number of states. Without magnetic field, the number of states in the dispersionless band in the sphere of the radius $k_{0}$ is equal to $\mathcal{N}=k_{0}^{3}/(6\,\pi^{2})$. In small magnetic fields, the number of states can be calculated using equation $bk_{0}(\sum_{n=0}^{n_{m}}\mathcal{G}(n)+1)/(2\pi^{2})\approx bk_{0}(x_{m}+3/2)/(2\pi^{2})$, where $\mathcal{G}(n)=1$ and the unity in the brackets accounts for additional midgap state for each $k_{z}$. The sum over $n$ was calculated using Eq. (7). Equating this result to $\mathcal{N}$, one obtains $x_{m}=k_{0}^{2}/(3\,b)-3/2$. Then the contribution from these bands is equal to

For the case of higher pseudospin, we calculated low-field susceptibility numerically. The smoothed thermodynamic potential was calculated using Euler–Maclaurin formula (7) and its field dependence was fitted by a second order polynomial in the low-field range below 1T. For $J=2$, corresponding dependence of susceptibility on the chemical potential at 50K is plotted on the left panel of the figure 2. The temperature dependences for several chemical potential values are plotted on the right panel of the figure 2.

If electron velocity $v$ and wave vector range $k_{0}$ are the same for different pseudospins, then the susceptibility increases with $J$ due to an increase of the number of energy branches and considered energy range. To facilitate comparison the susceptibility was calculated for scaled velocity $v=v_{0}/J$ so that the energy range $\epsilon_{0}$ was the same. Even in this case the magnitudes of extrema and the number of extrema in the dependence of $\chi(\mu)$ increases with $J$. For half-integer pseudospins the main extremum is a minimum and for integer pseudospins it is a maxumum due to the contribution of bounded middle bands. The temperature dependences of $\chi$ for integer pseudospins also exhibit maxima below 50K. In general, when chemical potential shifts from the energy of band-crossing point, $\chi$ becomes smaller. If the chemical potential reaches the next extremum, the susceptibility can still be quite large, but of the opposite sign, as shown by dashed lines in the right panel of the figure 2 for $J=3/2$ and 2.

Large values and non-monotonic dependence of orbital susceptibility on temperature and chemical potential were observed in several meterials with linear energy dispersion. For example, the minimum in the temperature dependence of magnetic susceptibility was observed in Weyl semimetal TaAs[23] at about 185K. Large values of diamagnetic susceptibility in graphene were discussed in [35] and were explained by the contribution of virtual interband transitions.

Another unusual feature, the change of the sign of orbital susceptibility, was also discussed in the literature. For example, the change of the sign of orbital susceptibility was considered in several two-dimensional lattices. In [36] it was shown that orbital susceptibility is always positive (paramagnetic) if the chemical potential is close to a saddle point of the band structure. The change of the sign of $\chi$ from negative to positive during transition from graphene to the dice lattice was considered in [37].

As an application to the real material, we calculated the contribution to low-field magnetic susceptibility from the states near the $\Gamma$ point in CoSi. Without spin-orbit coupling these states can be described by the Hamiltonian (1) for the pseudospin $J=1$. The magnetic susceptibility for this case was calculated in [32]. Its dependence on chemical potential and temperature is similar to that shown in the figure 2 for $J=1$ and differ only by a factor of two due to spin degeneracy.

In order to account for spin-orbit coupling, it is necessary to expand the basis set to include spin $|J=1,j\rangle\otimes|\uparrow\rangle,|J=1,j\rangle\otimes|\downarrow\rangle,j=1,0,-1$. The calculations showed that spin-orbit coupling in the zeroth order with respect to $\mathbf{k}$ can be written as $\hat{H}^{SOC}=2\,\Delta\,\hat{\mathbf{J}}\cdot\hat{\mathbf{s}}$, where $\hat{\mathbf{s}}$ is the operator of electron spin and the parameter $\Delta$ determines the magnitude of splitting. First-principle calculations yield $\Delta=18$ meV[8]. With the account of SOC, the 6-fold degenerate level at the $\Gamma$ point splits into a 4-fold degenerate level and a doublet shifted in energy by $\Delta$ and $-2\Delta$, respectively. Near the $\Gamma$ point the Hamiltonian has the form $\hat{H}=\hat{H}_{J=1}\otimes\hat{1}_{2\times 2}+\hat{H}^{SOC}$. Zeeman splitting can be taken into account by additional term $\hat{H}^{(Z)}=\mu_{B}B\hat{1}_{3\times 3}\otimes\hat{\sigma}_{3}$, where $\mu_{B}$ - Bohr magneton and $\hat{\sigma}_{3}$ - Pauli matrix.

For considered form of the spin-orbit coupling operator, electronic spectrum in magnetic field can be obtained for any $J$ similarly to described above. $\hat{\mathcal{H}}$ will be of five-diagonal form with matrix elements $\mathcal{H}_{j\sigma;j\sigma}=\hbar\,v\,k_{z}j+\mu_{B}\,B+2\,\Delta\,j\,\sigma$, $\mathcal{H}_{j\sigma;j+1,\,\sigma}=\mathcal{H}^{*}_{j+1,\,\sigma;j,\,\sigma}=i\,\hbar\,v\,(\hat{J}_{+})_{j,j+1}((n+\nu_{j,\,\sigma})\,b/2)^{1/2}$, and $\mathcal{H}_{j\sigma;j+1,\,\sigma-1}=\mathcal{H}^{*}_{j+1,\,\sigma-1;j,\,\sigma}=\Delta\,(\hat{J}_{+})_{j,j+1}(\hat{s}_{-})_{\sigma,\sigma-1}$, where $\nu_{j,\,\sigma}=J-j+s-\sigma$ and $(\hat{s}_{\pm})_{\sigma,\sigma\pm 1}=(s(s+1)-\sigma(\sigma\pm 1))^{1/2}$ are matrix elements of spin rising and lowering operators. Here, $s=1/2$ is the spin of electron, and $\sigma=1/2,-1/2$ enumerates its projections on $z$-axis. For given $n\geq 0$, the solution of eigenvalue problem for $\hat{\mathcal{H}}$ will give $2(2J+1)$ energy levels $\epsilon_{m,\,\gamma}(n,k_{z},b)$ and corresponding sets of expansion coefficients for $(m,\,\gamma)=(J,1/2),\,(J,-1/2),...,(-J,1/2),\,(-J,-1/2)$. Additional midgap solutions can be obtained for each $n=-(2J+1),...,-1$. Similarly to above consideration, the coefficients $c_{j,\,\sigma;\,n+\nu_{j,\,\sigma}}=0$, when $n+\nu_{j,\,\sigma}<0$.

The spectrum for CoSi near the $\Gamma$ point in magnetic field with the account of spin-orbit splitting is plotted in the figure 3. The band structure resembles the combination of Landau levels for $J=3/2$ and $1/2$-fermions shifted by $\Delta$ and $-2\Delta$ and distorted due to SOC. If the strength of SOC $\Delta$ tend to zero, the band structure will have the form shown in the figure 1 (left panel) for pseudospin $J=1$.

The dependences of low-field magnetic susceptibility on the chemical potential for these states are plotted in the figure 4. The susceptibility for this case is plotted with blue solid curve. For comparison the same calculation without Zeeman splitting is plotted with blue dotted curve. It corresponds to the contribution of the orbital part. The susceptibility without SOC and Zeeman splitting (doubled due to spin degeneracy) is plotted with the red dashed curve. It can be seen that the magnitudes of $\chi$ are similar for all considered cases. In the case of finite SOC strength, there are two bands with the bounded spectrum corresponding to two overlapping peaks in the figure.

In stoichiometric CoSi, the chemical potential is very close to the band crossing point and is equal to $-0.003$eV. The temperature dependence of the susceptibility for this case is expected to be similar to plotted in the figure 2 (right panel) for $J=1$ up to a factor of 2. The measurements of magnetic susceptibility in CoSi [25] showed that at high temperatures it is diamagnetic (negative) but at low temperatures it increases and at about 25K it changes its sign to positive in clean samples. The maximum positive values of $\chi$ were about $10^{-5}$emu/mol or $7.5\cdot 10^{-7}$emu/cm³. For less clean sample, the sign of susceptibility did not change with temperature, which can be caused, for example, by the deviation from stoichiometry. In this case, the chemical potential differs significantly from zero and the positive contribution to magnetic susceptibility decreases. It would be interesting in the future to consider the contribution to $\chi$ from the other parts of the spectrum to verify this assumption.

In this work we calculated low-field orbital magnetic susceptibility $\chi$ for multifold fermions with pseudospin $J>1$. Its dependence on temperature $T$ and chemical potential $\mu$ was considered. It was shown that $\chi(\mu)$ is nonmonotonic and demonstrate several extrema when $\mu$ approaches the energy of band crossing. The number of extrema increases with $J$. When $J$ is half-integer, the main extremum is diamagnetic (negative) and it is a minimum, while for integer pseudospin $\chi$ is paramagnetic (positive) and corresponds to a maximum. The temperature dependence of susceptibility also shows nonmonotonic behaviour. The calculations were extended to include spin-orbital coupling and illustrated by considering the contribution to susceptibility from multifold fermions near the $\Gamma$ point in CoSi.

We are grateful to Dr. Y.V. Ivanov for valuable comments and stimulating discussions.