--- title: 'Bloch State Tomography: Methods and Insights' url: https://www.emergentmind.com/topics/bloch-state-tomography type: topic --- # Bloch State Tomography: Methods and Insights Bloch state tomography denotes reconstruction protocols in which an unknown quantum object is represented by Bloch coordinates and inferred from measurement data. For a qubit, the target state is commonly written as \[ \rho=\frac12\bigl(I+r_x\sigma_x+r_y\sigma_y+r_z\sigma_z\bigr) \] or, equivalently, as \[ \rho(\xi)=\frac12\bigl(I+\xi_1\sigma_1+\xi_2\sigma_2+\xi_3\sigma_3\bigr), \] with physically allowed region given by the Bloch ball $\|\vec r\|\le1$; for a two-band Bloch problem, the cell-periodic eigenstate can be written relative to a reference basis as \[ u_k^1=\cos(\theta_k/2)\,|1\rangle+\sin(\theta_k/2)e^{i\phi_k}|2\rangle. \] In these settings, reconstructed Bloch data are used either to obtain a physical density matrix or to determine Berry curvature, the quantum metric tensor, Chern numbers, Wilson–Zak-loop eigenvalues, and $\mathbb Z_2$ invariants [1608.07983], [1509.02185], [2301.06090]. ## 1. State spaces, parametrizations, and measured quantities In single-qubit tomography, the Bloch or Stokes coordinates are expectation values of Pauli observables. If Pauli-$k$ is measured on a qubit state $\rho=\tfrac12(I+\vec r\cdot\vec\sigma)$, the two outcomes occur with probabilities \[ p_k(+1)=\tfrac12(1+r_k),\qquad p_k(-1)=\tfrac12(1-r_k). \] For conventional qubit state tomography, one measures each Pauli $\sigma_i$ exactly $N$ times; if $n_i^+$ and $n_i^-$ are observed, the empirical Stokes vector is \[ \xi^{(\mathrm{emp})}=\Bigl(\frac{n_1^+-n_1^-}{N},\frac{n_2^+-n_2^-}{N},\frac{n_3^+-n_3^-}{N}\Bigr), \] and this empirical vector may lie outside the Bloch ball [1608.07983]. The same Bloch-vector formalism underlies constrained hardware realizations. In a singlet–triplet double quantum dot, prepared states and effective measurements are parameterized as \[ \rho_i=\frac12\bigl[I+\mathbf r^{(i)}\cdot\boldsymbol\sigma\bigr],\qquad E_j=\lambda_j I+\mathbf R^{(j)}\cdot\boldsymbol\sigma, \] so that the Born probabilities are linear in Bloch components: \[ p_{j|i}=\mathrm{Tr}[E_j\rho_i]=\lambda_j+\sum_{a=x,y,z}R_a^{(j)}r_a^{(i)}. \] This formulation allows stochastic and systematic errors to be incorporated directly into the reconstructed state and measurement model [1306.1013]. In Bloch-band tomography, the reconstructed object is not a single density operator but the cell-periodic Bloch function over the Brillouin zone. For the lowest band in a two-band manifold one writes \[ u_k^1=\cos(\theta_k/2)\,|1\rangle+\sin(\theta_k/2)e^{i\phi_k}|2\rangle, \] where the two real angles $\theta_k,\phi_k$ parametrize the state up to a global gauge. The measurement task is then to recover amplitudes and phases of $u_k^1$ as functions of quasimomentum, rather than only three expectation values at one point in Hilbert space [1509.02185]. A common source of confusion is that “Bloch state tomography” is used in both senses. In qubit work it reconstructs a state inside the Bloch ball; in condensed-matter and cold-atom work it reconstructs momentum-dependent Bloch eigenfunctions. The shared structure is the use of Bloch coordinates as experimentally accessible parameters. ## 2. Information-geometric reconstruction of qubit states For conventional Pauli tomography, the joint model of outcomes is the product distribution \[ p_{\xi}(x_1,x_2,x_3)=\prod_{i=1}^3 \frac12(1+\xi_i x_i),\qquad x_i=\pm1. \] On the open cube $(-1,1)^3$, the Fisher information metric of this model is \[ g_{ij}(\xi)=\frac{\delta_{ij}}{1-\xi_i^2}, \] so the Stokes-parameter space is treated as a Riemannian manifold endowed with a dual-flat statistical-manifold structure [1608.07983]. The maximum-likelihood estimate is obtained by minimizing the Kullback–Leibler divergence from the empirical distribution: \[ \xi^{(\mathrm{MLE})}=\arg\min_{\|\xi\|\le1}D(\hat q\|p_\xi). \] When the empirical vector already satisfies $\|\xi^{(\mathrm{emp})}\|\le1$, the maximum-likelihood estimate coincides with the empirical estimate. When the empirical vector lies outside the Bloch ball, the maximum-likelihood estimate is the unique point on the boundary sphere that is orthogonal, with respect to the Fisher metric, to the geodesic joining the empirical point to the sphere. In this formulation, data processing based on the maximum-likelihood method is an orthogonal projection onto the Bloch sphere with respect to the Fisher metric rather than a Euclidean radial projection [1608.07983]. For randomized tomography with measurement counts $N_i$ and weights $\hat s_i=N_i/\sum_j N_j$, the metric becomes \[ g_{ii}(\xi)=\frac{\hat s_i}{1-\xi_i^2},\qquad g_{ij}=0\ (i\neq j). \] If the raw estimate is outside the physical region, the reconstruction reduces to solving \[ \xi_i(1-\xi_i^2)=\hat s_i\,\frac{\xi_i^{(\mathrm{emp})}-\xi_i}{\lambda}, \qquad \sum_{i=1}^3 \xi_i^2=1, \] for a Lagrange multiplier $\lambda>0$. The proposed algorithm transforms this into a single-variable root-finding problem \[ F(\lambda)=\sum_{i=1}^3 f_i(\lambda)^2-1=0, \] which can be solved by Newton–Raphson. The reported convergence is quadratic from a reasonable initial guess, typically in $5$–$10$ iterations, and the overall cost is $O(1)$ independent of sample size [1608.07983]. The statistical interpretation is explicit. The Fisher metric stretches directions with large variance and compresses those with small variance, so the correction respects the statistical reliability of each coordinate. The method guarantees a physical density matrix, is statistically consistent and efficient, achieves the Cramér–Rao bound asymptotically, is unbiased in the large-$N$ limit, and minimizes the KL divergence to the empirical distribution [1608.07983]. ## 3. Experimental architectures for qubit Bloch tomography Qubit implementations differ primarily in how Bloch components are mapped onto experimentally accessible observables. The reconstruction formulas remain Bloch-linear, but the measurement primitives range from projective Pauli readout to reflection spectroscopy, weak pointer shifts, and continuous Rabi scans. | Platform | Measurement primitive | Reconstruction relation | |---|---|---| | Pauli tomography | $X,Y,Z$ projective counts | $P_i(\pm)=\tfrac12(1\pm r_i)$ | | Singlet–triplet qubit | Native $z$ readout plus control pulses | $p_{j|i}=\lambda_j+\sum_a R_a^{(j)}r_a^{(i)}$ | | Kerr parametric oscillator | Reflection at a resonant line | $R(\omega_{20})\leftrightarrow \langle\sigma_z\rangle$ | | NV-center RQST | X–Rabi and Y–Rabi scans | $M_x(\theta)=r_z\cos\theta+r_y\sin\theta$, $M_y(\theta)=r_z\cos\theta-r_x\sin\theta$ | | Weak-measurement protocol | Three simultaneous weak couplings | $\delta q_i=g_i r_i$ | For the singlet–triplet qubit in a double quantum dot, initialization and readout are native only in the computational $z$ basis, identified approximately with $\{|S\rangle,|T_0\rangle\}$. Additional effective $x$- and $y$-measurements are generated either by preparing approximate eigenstates of those axes or by rotating an unknown state onto the $z$ axis before native readout, using the control Hamiltonian \[ H(\epsilon)=J(\epsilon)\sigma_z+\omega\sigma_x. \] A minimal tomographically complete set requires at least four independent states and three independent two-outcome measurements. The cited procedure also gives a self-consistent treatment of SPAM errors, free precession under a Lindblad model, and joint fitting of the unknown state and POVM parameters [1306.1013]. For Kerr parametric oscillators encoded in coherent states $|+\alpha\rangle$ and $|-\alpha\rangle$, a weak single-photon probe and reflection measurement provide a one-to-one mapping from the reflection coefficient to the population difference $\rho_{00}-\rho_{11}=\langle\sigma_z\rangle$ when the probe is tuned to a dominant transition. Pre-rotations $R_y(-\pi/2)$ and $R_x(+\pi/2)$ map $\langle\sigma_x\rangle$ and $\langle\sigma_y\rangle$ onto the same $\sigma_z$ readout channel, leading to \[ \rho_{\mathrm{tomo}}=\frac12[I+x\sigma_x+y\sigma_y+z\sigma_z]. \] The reported simulations give average tomography fidelity $\gtrsim97\%$ over the Bloch sphere, and $\gtrsim99\%$ with optimized ramp shapes and maximum-likelihood post-processing [2212.14627]. Rabi-based quantum state tomography replaces discrete calibration pulses by continuous X–Rabi and Y–Rabi scans. After fitting photoluminescence traces to decaying sinusoids, the Bloch components are obtained either from amplitudes or directly from fitted phases: \[ r_x=-A_y\sin\beta,\qquad r_y=A_x\sin\alpha,\qquad r_z=A_x\cos\alpha=A_y\cos\beta. \] For the NV-center electron spin, the reported average fidelity is $0.995$ over more than $40$ measurements on different Bloch-sphere states, with a maximum fidelity of $0.99992$; the same method is extended to the dark $^{14}\mathrm N$ nuclear spin via conditional gates and electron-spin readout [2408.13349]. Weak-measurement tomography provides a different limit. With three independent pointer couplings \[ H=\delta(t-t_0)\,[g_x\,\sigma_x\otimes p_x+g_y\,\sigma_y\otimes p_y+g_z\,\sigma_z\otimes p_z], \] and no post-selection discard, the first-order pointer shifts satisfy \[ \delta q_i=g_i\,r_i,\qquad \delta p_i=0. \] All three Bloch coordinates are thus extracted from one joint weak interaction, and the variance obeys \[ \mathrm{Var}(r_i)=\frac{\mathrm{Var}(q_i)}{N g_i^2}. \] This scheme is explicitly described as using all data rather than retaining only a post-selected subensemble [1212.3655]. These examples show that restriction to a single native basis does not preclude full tomography. The essential requirement is a calibrated mapping from measured signals to Bloch coordinates. ## 4. Bloch-band tomography and reciprocal-space geometry In Bloch-band systems, tomography reconstructs quasimomentum-dependent cell-periodic eigenfunctions. Under a constant external force $F$, the quasimomentum evolves as \[ k(t)=k(0)+Ft/\hbar, \] and, in the limit of an infinitely large force, the band evolution becomes purely geometric. The corresponding Wilson-line operator is \[ W_{k_i\to k_f}=\mathcal P\exp\!\Bigl[i\int_{k_i}^{k_f}A(k)\cdot dk\Bigr], \] with $A_{mn}(k)=i\langle u_k^m|\nabla_k u_k^n\rangle$. In a lattice whose relevant bands span the same two-dimensional Hilbert space at all $k$, this simplifies to \[ W_{k_i\to k_f}=e^{i(k_f-k_i)\cdot r}, \qquad W^{mn}_{k_i\to k_f}=\langle u_{k_f}^m|u_{k_i}^n\rangle. \] The Wilson line therefore encodes both intra-band Berry phases and interband couplings [1509.02185]. The tomography protocol in the honeycomb-lattice experiment reconstructs the lowest-band state at each $k$ through two measurements. First, a strong-force transport from a reference point $Q$ to $k_f$ yields the band-1 population \[ P_1=|W^{11}_{Q\to k_f}|^2=|\langle u_{k_f}^1|u_Q^1\rangle|^2=\cos^2(\theta_{k_f}/2), \] which determines $\theta_k$. Second, a Ramsey-type interferometer in quasimomentum space yields the phase \[ \phi_k=\mathrm{Arg}[W^{11}_{Q\to k_f}]-\mathrm{Arg}[W^{12}_{Q\to k_f}], \] and hence the full spinor \[ u_k^1=\cos(\theta_k/2)\,u_Q^1+\sin(\theta_k/2)e^{i\phi_k}u_Q^2. \] From the reconstructed wave functions, one computes Berry curvature, Chern numbers, Wilson–loop spectra, and, in multiband generalizations, $\mathbb Z_2$ indices [1509.02185]. A distinct implementation in a two-dimensional optical Raman lattice performs a complete Bloch-state tomography by direct spin-resolved measurements. The eigenstate in the first Brillouin zone is restricted to \[ |\Psi(q)\rangle=c_\uparrow(q)|q,\uparrow\rangle+c_\downarrow(q)|q+k_0,\downarrow\rangle, \] so the two plane-wave states act as north and south poles of an effective Bloch sphere. Stern–Gerlach time-of-flight yields $\sigma_z(q)=|c_\uparrow|^2-|c_\downarrow|^2$, while a $\pi/2$ momentum-transferring Raman pulse with controlled phase $\Delta\phi$ gives \[ P(q,t_R,\Delta\phi)=\langle\sigma_z\rangle\cos(2\Omega_{R0}t_R) +\bigl(\langle\sigma_y\rangle\cos\Delta\phi+\langle\sigma_x\rangle\sin\Delta\phi\bigr)\sin(2\Omega_{R0}t_R). \] Choosing $t_R=\pi/(4\Omega_{R0})$ with $\Delta\phi=\pi/2$ or $0$ yields $\langle\sigma_x\rangle$ or $\langle\sigma_y\rangle$, respectively [2301.06090]. Once $\langle\sigma_i(q)\rangle$ are known, the reconstructed lowest-band spinor is obtained from \[ \theta(q)=\arccos[-\langle\sigma_z(q)\rangle],\qquad \phi(q)=\arg[\langle\sigma_x(q)\rangle+i\langle\sigma_y(q)\rangle]. \] The quantum geometric tensor is then \[ \chi_{\mu\nu}(q)=\langle \partial_\mu u|(1-|u\rangle\langle u|)|\partial_\nu u\rangle, \] with real part $g_{\mu\nu}(q)$ and imaginary part related to the Berry curvature by \[ \mathcal F_{\mu\nu}(q)=-2\,\mathrm{Im}\,\chi_{\mu\nu}(q). \] For the reported detuning $\delta=-0.2E_r$, numerical integration on a grid of approximately $12{,}000$ $q$-points gives \[ C=\frac{1}{2\pi}\int_{\mathrm{FBZ}}\mathcal F_{xy}(q)\,d^2q\approx-1.00\pm0.02. \] The same data were used to test the pointwise inequality \[ \mathcal F_{xy}(q)^2\le 4\,\det[g(q)] \] and the global bound \[ \mathrm{vol}_g\ge |C|. \] This is a direct example in which Bloch-state tomography accesses both topology and quantum geometry rather than only band populations [2301.06090]. The reciprocal-space version of Bloch-state tomography therefore differs conceptually from qubit tomography. The reconstructed object is a gauge-fixed wave function over the Brillouin zone, and geometric quantities are derived from momentum-space derivatives of that wave function. ## 5. Minimality, confidence regions, and scaling Minimal qubit tomography is achieved with four outcomes. The tetrahedron measurement, a qubit SIC-POVM, has elements \[ E_i=\frac14\bigl(I+\mathbf v_i\cdot\boldsymbol\sigma\bigr), \] where the four Bloch vectors $\mathbf v_i$ point to the vertices of a regular tetrahedron and satisfy \[ \mathbf v_i\cdot\mathbf v_j= \begin{cases} 1,& i=j,\\[2mm] -\tfrac13,& i\neq j. \end{cases} \] For a double-slit qubit of light, this minimal tomography can be implemented on a single detection plane. Detecting at four carefully chosen positions $\xi_i$ realizes the required projectors, and the reconstruction is \[ p_i=\frac14(1+\mathbf v_i\cdot\mathbf r),\qquad \mathbf r=3\sum_{i=1}^4 p_i\,\mathbf v_i. \] The reported smallest solution uses $\zeta_0\approx3.4678$ and $\delta\approx2.7644$ for the Gaussian-slit geometry [1703.04260]. Reliable estimation requires more than a point estimator. For single-qubit tomography, prior-free confidence regions can be constructed directly in Bloch space. With empirical estimate $\hat r$ from Pauli-axis counts, total number of copies $n$, and confidence level $1-\alpha$, one obtains the confidence region \[ \mathcal C_{1-\alpha}= \{\,r\in\mathbb R^3:\|r-\hat r\|\le\delta\,\}\cap\{\|r\|\le1\}, \] where \[ \delta^2=\frac{2}{n}\Bigl(\ln\frac{2}{\alpha}+2\ln\binom{2n+3}{3}\Bigr). \] This region contains the true Bloch vector with probability at least $1-\alpha$, independently of any prior assumption on the distribution of possible states [1108.5329]. At the opposite end of the design space are scalable many-qubit protocols. Fourier-style tomography from single-pulse X/Y Bloch rotations applies \[ U(\vec t)=\prod_{q=1}^N U_q(t_q) \] to an $N$-qubit register, then measures in the laboratory $Z$ basis. Because each off-diagonal density-matrix element acquires a unique multi-qubit beat frequency, the state is recovered by a discrete-Fourier inversion over the measured times. At the operating point \[ g_q=\sqrt2\,\nu_q, \] the total tomography variance obeys \[ \sum_{a,b}\mathrm{Var}(\rho_{ab})\sim \frac{5^N}{N_{\mathrm{tot}}}, \] while the purity estimator satisfies \[ (\Delta\mu)^2\sim \frac{7^N}{N_{\mathrm{tot}}^2}. \] The scaling is still exponential in $N$—an unavoidable fact of full QST—but the reported base is reduced from $6$ to $5$ for full tomography, and to $\sqrt7\approx2.65$ for purity estimation [2101.05860]. These results delimit three distinct desiderata: symmetry and minimality, as in tetrahedral SIC measurements; rigorous uncertainty quantification, as in confidence-region tomography; and reduced exponential prefactors for many-qubit reconstruction. ## 6. Temporal Bloch tomography and generalized state formalisms Bloch tomography has also been extended from states at one time to “states over time.” For a system of local dimension $d$ monitored at times $t_0