Supplementary Information

I Computational setup for electron and thermal properties

Figure S1: (a) Electron band structure of BaMnSb2 for PBE+$U$ and PBE+$U$+SOC. (b) The phonon dispersion with and without SOC calculated by Phonopy. (c) The lattice thermal conductivity with and without SOC computed by ShengBTE. (d) The thermal conductivity convergence test with respect to the neighbor cutoff in thirdorder expansion.

The three computational setups (PBE+$U$, and PBE+$U$+SOC) are described by in the Methodology section of main text. Figure S1(a) reveals the electron band structure comparison for PBE+$U$ and PBE+$U$+SOC. The effects of SOC is mainly on the topological states on the Sb layer. In general, the band structures resemble each other, so we expect small difference in thermal property calculations. Figure S1(b) and (c) shows the phonon dispersion and lattice thermal conductivity with and without SOC. The lattice thermal conductivity has 724 finite-displacement jobs under five-neighbor cutoff. The negligible effect of SOC on phonon dispersion and thermal conductivity shows that PBE+$U$ is good enough to capture the thermal properties of BaMnSb2. Figure S1(d) shows the convergence of thermal conductivity in different directions ($\kappa^{a},\kappa^{b}$ and $\kappa^{c}$) with respect to the number of neighbors in the third order cutoff. With nine neighbors, the cutoff radius is 4.6 Å, and further increasing neighbor cutoff has less than 0.25 $\mathrm{Wm^{-1}K^{-1}}$ difference in the thermal conductivity.

Figure S2: (a) Three-phonon scattering rate of BaMnSb2 at 300 K with the frequency-squared trend indicated by the orange $\omega^{2}$ line. (b) The cumulative phonon thermal conductivity $\kappa_{\mathrm{ph}}$ as a function of mean free path at 300 K.

Figure S2(a) shows the scattering rate at 300 K obtained from ShengBTE. It is generally expected that, for low-frequency acoustic phonons, the scattering rate exhibits the frequency-squared trend, as delineated by the orange $\omega^{2}$ line. [1] The low-frequency scattering rates are higher than this trend, meaning the optical phonons strongly scatter the acoustic phonons, greatly reducing the lattice thermal conductivity. As a result, BaMnSb2 has a remarkably low lattice thermal conductivity, given a simple crystalline structure.

Figure S2(b) displays the size dependence of phonon thermal conductivity using ShengBTE by plotting the cumulative phonon thermal conductivity as a function of the mean free path at 300 K. The accumulation function of $\kappa_{\mathrm{ph}}$ implies how phonons with different mean-free-paths contribute to the overall phonon thermal conductivity. For BaMnSb2, its thermal conductivity is predominantly contributed by phonons with the mean free path around 10 nm, and reaches saturation when the sample size is roughly 100 nm. Using the phonon mean free path, we can identify which phonon-scattering mechanism, like phonon-phonon or phonon-boundary, becomes dominant for heat conduction at different size scale.

Figure S3: (a) Electrical conductivity calculated by BoltzTraP with constant relaxation time $\tau=1\times 10^{14}$ s. (b) Seebeck coefficient that is in the Fig. 3 of the main text. The negative slope of electrical conductivity corresponds to the sign of the Seebeck coefficient.

The Seebeck coefficient is calculated by the Mott formula, in which $S(E_{F})\sim-\frac{\sigma^{\prime}(E_{F})}{\sigma(E_{F})}$. [2] $\sigma(E_{F})$ is the electrical conductivity at Fermi level $E_{F}$. Figure S3(a) shows the numerically calculated $\sigma(E_{F})$ with the constant relaxation time $\tau=1\times 10^{14}$ s. In the conduction band $E_{F}>0$, we have $\sigma_{x}>\sigma_{y}>\sigma_{z}$ because the electron group velocity $v=\frac{\partial\varepsilon}{\partial k}$ evaluated in the electron band structure is the largest in $x$ direction $(v_{x}>v_{y}>v_{z})$. Even though, the density of states for conduction band is super low, the large group velocity $v_{x}$ brings in a steep increase, inducing a comparable $\sigma_{x}$ for conduction band and valence band with constant $\tau$. For the $z$ direction, $\sigma_{z}$ is drastically different for $E_{F}<0$ and $E_{F}>0$ due to the tiny $v_{z}$.

In Fig. S3(b), the sign of the Seebeck coefficient corresponds to the negative slope of $\sigma(E_{F})$. At low temperature (10 K), the change of slope happens near the neutral point $E_{F}=0$. At high temperature (100 K), $\sigma(E_{F})$ is smeared, and the slope change depends on the relative difference of $\sigma(E_{F}<0)$ and $\sigma(E_{F}>0)$. For $\sigma_{z}$, the large slope gives a large positive $S_{z}$, which is hole type conduction. For in-plane $\sigma_{x}$ and $\sigma_{y}$, the higher electron mobility results in smaller $\sigma$ difference, and the Seebeck positive-to-negative transition is still close to the neutral point $E_{F}=0$. Therefore, the in-plane conduction is electron type, and BaMnSb2 has applications as a goniopolar material.

II Quantum oscillation for Fermi level estimation

Figure S4: (a) Electron Fermi surface at Fermi level 35 meV above the neutral point. (b) Hole Fermi surface at Fermi level $-$5 meV below the neutral point. (c) Quantum oscillation frequency at a function of Fermi level. The small positive Fermi level has a cylinder shape, while larger Fermi level has a barrel shape with a bulge at the center. The barrel-shape Feri surface has two extremal orbitals perpendicular to magnetic field direction $B$, giving rise to the larger blue and smaller orange areas or frequencies. (d) The comparison of computed and experimental quantum oscillation frequencies. The computational data varies with magnetic field rotating from $z$ to $x$ direction. The two barrel shapes describe how the extremal orbitals evolve with different magnetic field direction. The experimental data is from the Shubnikov-de Haas (SdH) oscillations of Ref. 3.

Quantum oscillation measurements are useful tools to determine the Fermi surfaces of the material. [2] For metallic systems, applying a strong magnetic field (B) will lead to the Landau level quantization of electron orbitals, and thus the carrier density of states oscillates with the B field, and oscillations manifest in many experimental observations such as the Shubnikov-de Haas (SdH) oscillations in Ref. 3. The oscillation frequency $F$ is closely related to the shape of Fermi surface by the Onsager relation $F=\frac{\hbar}{2\pi e}A$. $F$ is proportional to extremal areas $A$ of the Fermi surface perpendicular to the applied magnetic field, where $\hbar$ is the reduced Planck’s constant and $e$ is the electron charge.

We can evaluate the oscillation frequency by a fortran-based package SKEAF [4]. With a particular shape of Fermi surface like Fig. S4(a) and (b), slices of Fermi surface perpendicular to the B field are first generated, and the areas of the Fermi surface slices are computed to find the extremal values. Figure S4(c) shows the calculated oscillation frequencies as a function of Fermi level. The inside cylinder and barrel are schematic illustrations of the Fermi surface with smaller and larger Fermi levels. With magnetic field in $z$ direction, the extremal orbitals are the blue and orange dashed circles. The blue (orange) electron orbital has a larger (smaller) area, and thus corresponds to a larger (smaller) oscillation frequency in the blue (orange) line. Figure S4(d) plots the oscillation frequency with rotating B field direction from $z$ to $x$. With a larger tilting angle, the extremal orbitals have larger areas. We have found the oscillation frequency near Fermi level 35 meV agree reasonably well with the experimental data[3]. Hence, the Fermi level position is estimated to be 35 meV.