What Is an Outcome

An outcome is the label of a recorded event in a specified measurement procedure. Its meaning comes from the apparatus calibration, resolution and rule for grouping records. A probability, an expectation value and a state-vector component are not themselves an individual outcome. The outcome problem asks how a physical theory represents the occurrence of such a record.

Labels and physical records

A detector may produce a voltage trace, a click time or an image. Turning that record into a mathematical label requires a rule: thresholds, time bins, image regions or another declared classification. Quantum measurement theory assigns an effect $E_a$ to each label and uses $p(a)=\operatorname{Tr}(\rho E_a)$.

Changing the labels without changing the physical experiment is coarse-graining. If labels $a$ are grouped into an event $C$, its effect and probability are

$$E_C=\sum_{a\in C}E_a,\qquad p(C)=\sum_{a\in C}p(a).$$

This is an operational event algebra. It does not license combining outcomes from incompatible procedures into one joint distribution of pre-existing values.

An expectation may never occur

For a spin measurement with reported values $\pm\hbar/2$ and equal probabilities, the expectation is zero. No individual ideal outcome is zero. Similarly, a weak pointer displacement proportional to a weak value is a feature of an ensemble protocol, not automatically an eigenvalue possessed by one system. The weak-value bridge explains that distinction.

For an absolutely continuous outcome distribution, probabilities belong to ranges and each exact point has zero probability. A continuum-valued measurement can also have discrete atoms, so a continuous outcome space alone does not imply zero point probabilities. Finite detector resolution makes the recorded bins central to comparison with data. The outcome alphabet should therefore be specified at the level at which the experiment actually records information.

Definiteness has a scope

Operational use starts from recorded labels and predicts their correlations. An interpretation then asks whether a label is globally unique, branch-relative, relative to an interaction, or represented by additional variables. These alternatives concern the physical status of records, not whether the symbols in a data file are well-defined.

A measurement can also fail to produce a usable registered event. A no-click or abort may be a legitimate outcome that must be included in the statistical model. Discarding it after seeing other data is not innocent relabeling; it is postselection and can change inferred correlations.

Exercise: binning a detector

A detector has four labels with probabilities $(0.1,0.2,0.3,0.4)$. An analysis reports only even versus odd labels, indexed from zero. Find the reported probabilities. Is this equivalent to keeping only even labels?

Solution

Even labels zero and two have total probability $0.4$; odd labels one and three have probability $0.6$. Both coarse outcomes remain. Keeping only even labels instead conditions on an event of probability $0.4$ and renormalizes the retained records to probabilities $0.25,0.75$. Coarse-graining preserves the full trial population; postselection changes it.

References

• Busch, Paul, Pekka J. Lahti, and Peter Mittelstaedt, The Quantum Theory of Measurement, 2nd edition, Springer (1996), doi:10.1007/978-3-540-37205-9.
• Maudlin, Tim, “Three Measurement Problems,” Topoi 14, 7–15 (1995), doi:10.1007/BF00763473 — the distinct issue of definite outcomes.