Simple Algorithm
Quantum Computing · Rydberg State · 28 min

Qudits & the Generalized Bloch Picture

Gell-Mann matrices and why higher dimensions are not balls.

For a qubit, every unit vector in R3\R^3 is a pure state and every point of the ball is a state. For a qutrit the state space lives in R8\R^8, and it is not a ball: in some directions it ends at radius 12\tfrac12, and the antipode of a pure state is not a state at all. This lesson builds the generalized Bloch vector and maps the convex body it fills.

What you’ll learn

  • Expand any qudit density matrix in the generalized Gell-Mann basis and derive the purity formula Tr⁡ρ2=1d(1+(d−1)∣r⃗∣2)\Tr\rho^2=\tfrac1d\big(1+(d-1)\abs{\vec r}^2\big).
  • Prove that for d≥3d\ge3 the Bloch body lies between the balls of radius 1d−1\tfrac1{d-1} and 11, touches both, and is not centrally symmetric.
  • Count dimensions: pure states form a 2(d−1)2(d-1)-dimensional subset of the (d2−2)(d^2-2)-sphere, and unitaries form a small subgroup of the rotations.
  • Use the Weyl–Heisenberg (clock and shift) operators as the qudit analogue of the Paulis, and describe open problems such as SIC existence.

The generalized Bloch vector

DerivationBloch vector and purity
  1. ρ=1d(I+κd r⃗⋅λ⃗),κd=d(d−1)2,r⃗∈Rd2−1\rho=\frac1d\Big(\id+\kappa_d\,\vec r\cdot\vec\lambda\Big),\qquad\kappa_d=\sqrt{\tfrac{d(d-1)}{2}},\qquad\vec r\in\R^{d^2-1}

    Expand the unit-trace Hermitian matrix ρ\rho in {I,λa}\{\id,\lambda_a\}. The factor κd\kappa_d is chosen so pure states have ∣r⃗∣=1\abs{\vec r}=1; for d=2d=2, κ2=1\kappa_2=1 recovers 12(I+r⃗⋅σ⃗)\tfrac12(\id+\vec r\cdot\vec\sigma).

  2. ra=d2κdTr⁡(ρλa)=d2κd⟨λa⟩r_a=\frac{d}{2\kappa_d}\Tr(\rho\lambda_a)=\frac{d}{2\kappa_d}\expval{\lambda_a}

    Components are expectation values, exactly as for a qubit.

  3. Tr⁡ρ2=1d2(d+2κd2∣r⃗∣2)=1d(1+(d−1)∣r⃗∣2)\Tr\rho^2=\frac1{d^2}\Big(d+2\kappa_d^2\abs{\vec r}^2\Big)=\frac1d\Big(1+(d-1)\abs{\vec r}^2\Big)

    Use Tr⁡λa=0\Tr\lambda_a=0 and Tr⁡(λaλb)=2δab\Tr(\lambda_a\lambda_b)=2\delta_{ab}. The purity runs from 1d\tfrac1d at r⃗=0\vec r=0 (maximally mixed) to 11 at ∣r⃗∣=1\abs{\vec r}=1.

  4. Tr⁡(ρr⃗ ρs⃗)=1d(1+(d−1) r⃗⋅s⃗)\Tr(\rho_{\vec r}\,\rho_{\vec s})=\frac1d\Big(1+(d-1)\,\vec r\cdot\vec s\Big)

    Overlaps are again dot products. For pure states this is ∣⟨ϕ∣ψ⟩∣2\abs{\braket{\phi|\psi}}^2.

So far everything looks like the qubit. The difference is positivity. A density matrix must satisfy ρ≥0\rho\ge0, and for d≥3d\ge3 that is not implied by ∣r⃗∣≤1\abs{\vec r}\le1. A Hermitian unit-trace matrix with eigenvalues (23,23,−13)(\tfrac23,\tfrac23,-\tfrac13) has Tr⁡ρ2=1\Tr\rho^2=1, so its Bloch vector has length 1, but it is not a state.

Not a ball

DerivationProof
  1. ρr⃗≥0  ⟺  1+κd∣r⃗∣ μmin⁡(r^⋅λ⃗)≥0\rho_{\vec r}\ge0\iff1+\kappa_d\abs{\vec r}\,\mu_{\min}(\hat r\cdot\vec\lambda)\ge0

    ρ≥0\rho\ge0 means its smallest eigenvalue is non-negative. Write r⃗=∣r⃗∣r^\vec r=\abs{\vec r}\hat r; A=r^⋅λ⃗A=\hat r\cdot\vec\lambda is traceless Hermitian with Tr⁡A2=2\Tr A^2=2.

  2. μmin⁡(A)≥−2(d−1)d\mu_{\min}(A)\ge-\sqrt{\tfrac{2(d-1)}{d}}

    Let μ1=−a\mu_1=-a be the smallest eigenvalue. The other d−1d-1 eigenvalues sum to aa, so by Cauchy–Schwarz their squares sum to at least a2/(d−1)a^2/(d-1). Then 2≥a2(1+1d−1)2\ge a^2\big(1+\tfrac1{d-1}\big). Equality holds for the spectrum (−a,ad−1,…,ad−1)(-a,\tfrac a{d-1},\dots,\tfrac a{d-1}), i.e. A∝d ∣ψ⟩ ⁣⟨ψ∣−IA\propto d\,\dyad\psi-\id up to sign.

  3. ∣r⃗∣≤1κdd2(d−1)=1d−1  ⟹  ρr⃗≥0\abs{\vec r}\le\frac{1}{\kappa_d}\sqrt{\frac{d}{2(d-1)}}=\frac1{d-1}\;\Longrightarrow\;\rho_{\vec r}\ge0

    So every vector of length at most 1d−1\tfrac1{d-1} is a state, and the bound is tight in the extremal direction. There the boundary state is (I−∣ψ⟩ ⁣⟨ψ∣)/(d−1)(\id-\dyad\psi)/(d-1), the maximally mixed state on the complement of a pure state, sitting opposite to ∣ψ⟩\ket\psi.

  4. ∣r⃗∣≤1,∣r⃗∣=1  ⟺  ρ2=ρ\abs{\vec r}\le1,\qquad \abs{\vec r}=1\iff\rho^2=\rho

    The outer bound is Tr⁡ρ2≤1\Tr\rho^2\le1. A state has purity 1 iff it is pure, so the outer sphere meets Bd\mathcal B_d exactly in the pure states. □\square

Predict before you look

With pure qutrit states normalized to ∣r⃗∣=1\abs{\vec r}=1, what is the radius of the largest ball around r⃗=0\vec r=0 consisting entirely of valid states?

Show the answer

0.5

1d−1=12\tfrac1{d-1}=\tfrac12 for d=3d=3. The closest boundary points are states like 12(∣0⟩ ⁣⟨0∣+∣1⟩ ⁣⟨1∣)\tfrac12(\dyad0+\dyad1), half-way along an edge between two pure states.

Two qubit facts break. Orthogonality: Tr⁡(ρσ)=0\Tr(\rho\sigma)=0 for pure states means r⃗⋅s⃗=−1d−1\vec r\cdot\vec s=-\tfrac1{d-1}, so dd mutually orthogonal pure states form a regular simplex on the unit sphere, not a pair of antipodes. Antipodes: if r⃗\vec r is pure, then ρ−r⃗=2dI−∣ψ⟩ ⁣⟨ψ∣\rho_{-\vec r}=\tfrac2d\id-\dyad\psi has eigenvalue 2d−1<0\tfrac2d-1<0. Finally, which unit vectors are pure is itself a nonlinear condition. For d≥3d\ge3, r⃗\vec r is pure iff ∣r⃗∣=1\abs{\vec r}=1 and ra=κdd−2 dabc rbrcr_a=\frac{\kappa_d}{d-2}\,d_{abc}\,r_br_c, with dabc=14Tr⁡(λa{λb,λc})d_{abc}=\tfrac14\Tr(\lambda_a\acomm{\lambda_b}{\lambda_c}) (for d=3d=3 the prefactor is 3\sqrt3).

Check your understanding

Two orthogonal pure qutrit states have Bloch vectors separated by which angle?

  1. 90∘90^\circ
  2. 180∘180^\circ
  3. arccos⁡(−13)≈109.5∘\arccos(-\tfrac13)\approx109.5^\circ
  4. 120∘120^\circ
Show the answer

D. 120∘120^\circ

r⃗⋅s⃗=−1d−1=−12\vec r\cdot\vec s=-\tfrac1{d-1}=-\tfrac12, so the angle is 120∘120^\circ. The basis states ∣0⟩ ⁣⟨0∣,∣1⟩ ⁣⟨1∣,∣2⟩ ⁣⟨2∣\dyad0,\dyad1,\dyad2 form an equilateral triangle.

Cross-sections and counting

Different 2D slices of B3\mathcal B_3 through the origin look completely different:

  • The (λ3,λ8)(\lambda_3,\lambda_8) plane holds the diagonal states diag⁡(p0,p1,p2)\diag(p_0,p_1,p_2): a probability simplex. It is an equilateral triangle with vertices ∣0⟩ ⁣⟨0∣,∣1⟩ ⁣⟨1∣,∣2⟩ ⁣⟨2∣\dyad0,\dyad1,\dyad2 on the unit circle at (32,12)(\tfrac{\sqrt3}2,\tfrac12), (−32,12)(-\tfrac{\sqrt3}2,\tfrac12) and (0,−1)(0,-1). Its inscribed circle has radius 12\tfrac12, touching the edges at states like 12(∣0⟩ ⁣⟨0∣+∣1⟩ ⁣⟨1∣)\tfrac12(\dyad0+\dyad1).
  • The (λ1,λ2)(\lambda_1,\lambda_2) plane gives ρ\rho with eigenvalues 13(1±3∣r⃗∣)\tfrac13(1\pm\sqrt3\abs{\vec r}) and 13\tfrac13: a disk of radius 13≈0.58\tfrac1{\sqrt3}\approx0.58 containing no pure states.
ddBloch space d2−1d^2-1pure states 2(d−1)2(d-1)inner radius 1d−1\tfrac1{d-1}dim⁡PU(d)\dim PU(d)dim⁡SO(d2−1)\dim SO(d^2-1)
232133
38412\tfrac12828
415613\tfrac1315105
Pure states, CPd−1\C P^{d-1}, have real dimension 2(d−1)2(d-1), a thin subset of the (d2−2)(d^2-2)-dimensional unit sphere once d≥3d\ge3. Unitaries act on Bloch space as rotations, but they form only a (d2−1)(d^2-1)-dimensional subgroup of SO(d2−1)SO(d^2-1). For d=2d=2, and only then, every rotation of the ball is a unitary (PU(2)≅SO(3)PU(2)\cong SO(3)).

Qudit Paulis and the frontier

For computation the Gell-Mann basis is less useful than the Weyl–Heisenberg operators, the clock and shift. With ω=e2πi/d\omega=e^{2\pi i/d}, define Z∣j⟩=ωj∣j⟩Z\ket j=\omega^j\ket j and X∣j⟩=∣j+1 mod d⟩X\ket j=\ket{j+1\bmod d}.

ZX=ω XZ,Xd=Zd=I,Tr⁡ ⁣[(XaZb)†Xa′Zb′]=d δaa′δbb′ZX=\omega\,XZ,\qquad X^d=Z^d=\id,\qquad\Tr\!\big[(X^aZ^b)^\dagger X^{a'}Z^{b'}\big]=d\,\delta_{aa'}\delta_{bb'}
The d2d^2 unitaries XaZbX^aZ^b form an orthogonal operator basis, generalizing {I,X,Y,Z}\{\id,X,Y,Z\} (for d=2d=2, ω=−1\omega=-1 and Y∝XZY\propto XZ). They are unitary but, for d≥3d\ge3, not Hermitian.

The normalizer of this group is the qudit Clifford group. For prime dd the stabilizer formalism, stabilizer codes and the Gottesman–Knill simulation all carry over, and non-stabilizer magic states again supply universality.

Key takeaways

  • A qudit state is ρ=1d(I+d(d−1)/2  r⃗⋅λ⃗)\rho=\tfrac1d\big(\id+\sqrt{d(d-1)/2}\;\vec r\cdot\vec\lambda\big) with r⃗∈Rd2−1\vec r\in\R^{d^2-1} and purity 1d(1+(d−1)∣r⃗∣2)\tfrac1d\big(1+(d-1)\abs{\vec r}^2\big).
  • For d≥3d\ge3 the Bloch body is convex and lies between the balls of radius 1d−1\tfrac1{d-1} and 11, touching both; it is neither a ball nor centrally symmetric.
  • Pure states form a 2(d−1)2(d-1)-dimensional subset of the unit sphere; orthogonal pure states satisfy r⃗⋅s⃗=−1d−1\vec r\cdot\vec s=-\tfrac1{d-1}, and unitaries are a small subgroup of SO(d2−1)SO(d^2-1).
  • The Weyl–Heisenberg operators XaZbX^aZ^b replace the Paulis for qudit stabilizers, Cliffords and discrete Wigner functions; SIC existence (Zauner's conjecture) remains open.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

How many real parameters does a general qutrit density matrix have, and what is the real dimension of the set of pure qutrit states?

  1. 8 and 7
  2. 8 and 4
  3. 9 and 4
  4. 8 and 6
Show the answer

B. 8 and 4

d2−1=8d^2-1=8 Bloch components. Pure states form CP2\C P^2, of real dimension 2(d−1)=42(d-1)=4: a thin subset of the 7-sphere ∣r⃗∣=1\abs{\vec r}=1.

Question 2

For a pure qutrit state with Bloch vector r⃗\vec r, what is the point −r⃗-\vec r?

  1. The orthogonal pure state
  2. A mixed state
  3. The maximally mixed state
  4. Not a state: ρ−r⃗\rho_{-\vec r} has a negative eigenvalue
Show the answer

D. Not a state: ρ−r⃗\rho_{-\vec r} has a negative eigenvalue

ρ−r⃗=2dI−∣ψ⟩ ⁣⟨ψ∣\rho_{-\vec r}=\tfrac2d\id-\dyad\psi has eigenvalue 23−1=−13\tfrac23-1=-\tfrac13 on ∣ψ⟩\ket\psi. Along −r⃗-\vec r the Bloch body ends at distance 12\tfrac12, at (I−∣ψ⟩ ⁣⟨ψ∣)/2(\id-\dyad\psi)/2.

Question 3

Restrict qutrit Bloch vectors to the plane spanned by the λ3\lambda_3 and λ8\lambda_8 directions (diagonal density matrices). What shape do the valid states form?

  1. A disk of radius 1
  2. A disk of radius 12\tfrac12
  3. An equilateral triangle inscribed in the unit circle
  4. A square
Show the answer

C. An equilateral triangle inscribed in the unit circle

Diagonal states are probability vectors (p0,p1,p2)(p_0,p_1,p_2): a 2-simplex. Its vertices ∣0⟩ ⁣⟨0∣,∣1⟩ ⁣⟨1∣,∣2⟩ ⁣⟨2∣\dyad0,\dyad1,\dyad2 lie on the unit circle at 120∘120^\circ, and its inscribed circle has radius 12=1d−1\tfrac12=\tfrac1{d-1}.

Question 4

For the qudit clock and shift operators Z∣j⟩=ωj∣j⟩Z\ket j=\omega^j\ket j and X∣j⟩=∣j+1⟩X\ket j=\ket{j+1}, with ω=e2πi/d\omega=e^{2\pi i/d}, which relation holds?

  1. ZX=ω XZZX=\omega\,XZ
  2. XZ=ZXXZ=ZX
  3. ZX=−XZZX=-XZ for every dd
  4. X2=IX^2=\id
Show the answer

A. ZX=ω XZZX=\omega\,XZ

ZX∣j⟩=ωj+1∣j+1⟩ZX\ket j=\omega^{j+1}\ket{j+1} while XZ∣j⟩=ωj∣j+1⟩XZ\ket j=\omega^j\ket{j+1}. For d=2d=2, ω=−1\omega=-1 recovers the Pauli anticommutation ZX=−XZZX=-XZ.

Further reading

  • I. Bengtsson & K. Życzkowski, Geometry of Quantum States: An Introduction to Quantum Entanglement, 2nd ed. (Cambridge, 2017).
  • G. Kimura, The Bloch vector for N-level systems, Physics Letters A (2003).
  • M. S. Byrd & N. Khaneja, Characterization of the positivity of the density matrix in terms of the coherence vector representation, Physical Review A (2003).
  • J. M. Renes, R. Blume-Kohout, A. J. Scott & C. M. Caves, Symmetric informationally complete quantum measurements, Journal of Mathematical Physics (2004).
  • D. Gross, Hudson's theorem for finite-dimensional quantum systems, Journal of Mathematical Physics (2006).