A geometry for one qubit
Definition 1.2.1
(Qubit density matrix)
A qubit density matrix is a positive semidefinite Hermitian matrix of trace one.
The Pauli matrices
are Hermitian, traceless and satisfy . Every qubit density matrix has a unique expansion
Definition 1.2.2
(Bloch vector and Bloch ball)
The vector in the expansion above is the Bloch vector of . The set of all Bloch vectors of qubit states is the Bloch ball.
The Pauli multiplication identities give . Consequently the eigenvalues of are . Positivity holds exactly for : qubit states form the closed Bloch ball.
The purity is . Here . A state is pure exactly when , and the center represents . Interior points are mixed. This geometric picture is special to a single qubit; a general higher-dimensional state space is not simply a Euclidean ball of this form.
Worked example 1.2.3
(Bloch vectors of pure and mixed states)
For , , so its vector is . The equally weighted mixture of and is . So is the equally weighted mixture of and : different preparation ensembles can describe the same density matrix and give the same statistics for every measurement on that qubit.
From a measurement to a success probability
Suppose a source chooses or with equal probability. A two-outcome measurement is specified by positive semidefinite operators with ; on outcome we guess . Since , its success probability is
Any finite-outcome measurement and decision rule can be coarse-grained by summing the effects that lead to each guess; this gives the same success probability. So it is enough to optimize over two-outcome measurements, and the optimum includes generalized measurements, not merely a chosen computational-basis measurement.
Definition 1.2.4
(Trace norm)
For a square matrix , the trace norm is . For a Hermitian matrix it is the sum of the absolute values of its eigenvalues.
Theorem 1.2.5
(Optimal equal-prior discrimination of two states)
Let be density matrices, each chosen with probability , and let one copy be available. The maximum over all measurements and decision rules of the probability of guessing the label correctly is
It is attained by the projector onto the positive eigenspace of together with .
Proof
(Bounding the trace against the difference)
Write the Hermitian difference in an orthonormal eigenbasis , with eigenvalues . Since , each diagonal entry lies in . Therefore
because terms with are at most zero and on the terms with .
Equality holds when is the projector onto the span of the eigenvectors with and , since then exactly where and elsewhere. Zero eigenspaces can be assigned to either outcome without changing the optimum.
Let be the sum of the positive eigenvalues and the sum of the absolute values of the negative ones. Because , we have , while . Hence , and
Proof X-ray
Where each hypothesis enters
Three facts carry the argument, and each is used in one place. The identity together with turns the two-term success probability into a single trace . The bound is used only to force the diagonal entries into ; off-diagonal entries of never contribute, because is diagonal in the chosen basis. The one non-obvious move is that choice of basis: it reduces an optimization over operators to an optimization of weights , one per eigenvalue, where the best weight is for a positive eigenvalue and for a negative one. The argument closes because that weight pattern is the spectral projector, itself an admissible effect. Equal priors are needed for the final form: is what converts the positive part into half the trace norm. With unequal priors the operator has nonzero trace and the formula changes.
For qubits with Bloch vectors , the difference is , whose eigenvalues are . Hence and
Antipodal pure states are distinguishable with certainty; identical states give only the random-guess success probability . Intermediate distances give intermediate optimal success.
Worked example 1.2.6
(Distinguishing the states 0 and plus)
Distinguish from with equal priors. Their Bloch vectors are and , separated by . The optimal success probability is . Measuring in the computational basis and guessing on outcome succeeds with probability , which is valid but not optimal.
Worked example 1.2.7
(A pure state against the maximally mixed state)
Take and , with Bloch vectors and . Directly, , whose trace norm is , and the distance between the vectors is also . Thus . The optimal measurement is the computational-basis measurement, guessing on outcome : it is correct with probability if the state is and with probability if the state is , so the average is .
Worked example 1.2.8
(Two states symmetric about the center)
Take Bloch vectors and . The distance is , so . The difference is , so the optimal measurement is again in the computational basis. The two states lie inside the ball, so they are mixed, yet their distance alone fixes the optimum; the radii of the vectors play no further role.
Counterexample mode
A particular nonorthogonal pair cannot be identified perfectly
A tempting claim is that a sufficiently clever measurement can always tell two different pure states apart on every run. Test it on and . The overlap is nonzero, and the calculation above gives for a single copy with equal priors. A measurement can sometimes provide informative evidence, but no measurement in this setting guarantees the correct answer on every run. Perfect discrimination of pure states requires orthogonality, as for the antipodal pair . Repeating an experiment with independently prepared copies is a different resource scenario.
Common mistake
(Bloch angle is not Hilbert-space angle)
Orthogonal pure states are antipodal on the Bloch sphere, not at a right angle. The states and have Bloch vectors at angle , yet their overlap is , which corresponds to a Hilbert-space angle of . Reading a right angle on the sphere as orthogonality would suggest and can be told apart perfectly; the formula shows the optimal success is only about .
Common mistake
(Using the equal-prior formula with unequal priors)
The formula assumes each label has probability . If but the label has prior , the formula gives , yet always guessing succeeds with probability . With unequal priors the measurement must be tuned to the weighted operator .
Practice and solutions
In the checks below the two labels have equal prior probability and one copy is available.
Quick check
(Find the Bloch vector and purity of .)
Write in the form .
Solution · Answer
, so and the purity is . As a check, .
Quick check
(What is the optimal equal-prior success probability for distinguishing from ?)
Compare the Bloch vectors.
Solution · Answer
Their vectors are antipodal, at distance two, giving success . A computational-basis measurement attains it.
Quick check
(Two source labels both prepare . What is the best success probability based on one qubit?)
What is the trace norm of the difference?
Solution · Answer
The density matrices agree, so the difference is zero, its trace norm is zero, and the optimum is : the qubit carries no information about the label.
Quick check
(Find the optimal success probability for states with Bloch vectors and .)
Use the distance formula.
Solution · Answer
The distance is one, giving .