Holevo Bound (Overview)

How much classical information can be packed into, and reliably recovered from, a quantum system? A qubit has a continuum of possible states, which might suggest unlimited storage. The Holevo bound demolishes that hope: it caps the classical information accessible from a quantum ensemble at a value that, for a $d$ -level system, never exceeds $\log_2 d$ bits.

The setup

Alice encodes a classical symbol $x$ , drawn with probability $p_x$ , into a quantum state $\rho_x$ and sends it to Bob. Bob performs some measurement and obtains an outcome $y$ , learning about $x$ . The information Bob extracts is the accessible information $I(X\!:\!Y)$ — the classical mutual information between Alice's symbol and Bob's outcome, maximized over all of Bob's possible measurements.

The bound

The Holevo bound states that, for every measurement Bob could make,

\boxed{\,I(X\!:\!Y) \le \chi \equiv S\!\Big(\sum_x p_x \rho_x\Big) - \sum_x p_x\, S(\rho_x).\,}

The right-hand side $\chi$ is the Holevo quantity (or Holevo $\chi$ ). It is the entropy of the average state minus the average of the entropies — a quantum analogue of mutual information often written $\chi = S(\bar\rho) - \sum_x p_x S(\rho_x)$ with $\bar\rho = \sum_x p_x\rho_x$ .

Because $\chi \le S(\bar\rho) \le \log_2 d$ , the accessible information from a $d$ -dimensional system is at most $\log_2 d$ bits. One qubit carries at most one bit of recoverable classical information, despite its continuous state space.

Why it is true (sketch)

Introduce the classical register $X$ , the quantum message $Q$ , and Bob's measurement outcome $Y$ . The chain $X \to Q \to Y$ is a quantum-to-classical process, and the monotonicity of mutual information under such processing (a consequence of strong subadditivity) gives $I(X\!:\!Y) \le I(X\!:\!Q)$ . A direct computation identifies $I(X\!:\!Q)$ for the classical–quantum state $\rho_{XQ} = \sum_x p_x |x\rangle\langle x| \otimes \rho_x$ with exactly the Holevo quantity $\chi$ . The bound is therefore the data-processing inequality applied to encoding then measuring.

When the bound is tight

Orthogonal states. If the $\rho_x$ are mutually orthogonal pure states, Bob can distinguish them perfectly: $\chi = H(X)$ and $I(X\!:\!Y) = H(X)$ . Orthogonality is what makes information classical-like.
Non-orthogonal states. Overlapping states cannot be perfectly distinguished; $\chi < H(X)$ , and some information is irretrievably lost. This is why you cannot store two bits in one qubit by using four non-orthogonal states.

Consequences

The Holevo bound underlies the Holevo–Schumacher–Westmoreland (HSW) theorem, which establishes that the maximum of $\chi$ over input ensembles is the classical capacity of a quantum channel (per use, for product-state encodings). It also explains the impossibility of superdense coding beyond its known limit and frames the cost of communication in any quantum protocol.

The take-away: a quantum system's continuous state space does not buy extra classical storage. The ceiling is $\log_2 d$ bits — set by the same von Neumann entropy and subadditivity inequalities that govern the rest of the module. The next lesson turns to the quantum analogue of source coding, Schumacher compression.