Inner Products, Orthogonality, and Projection

Overlap is the whole game

The inner product answers a single question: how much does one state lean in the direction of another? For real vectors it is the dot product you already know. Quantum states carry complex amplitudes, so the first vector must be conjugated before anything is multiplied:

Skip the conjugation and you get answers that are wrong in a specific way: phases that should cancel do not, and probabilities come out complex or negative. With the conjugate in place, the result is always a complex number whose magnitude never exceeds the product of the two lengths.

When that number is zero, the states are orthogonal — they share no direction at all. Orthogonal states can be distinguished perfectly by a single measurement in the right basis. Non-orthogonal states cannot be, no matter how clever the instrument. That is a theorem, not an engineering limitation.

Projectors and the Born rule

Projection onto a normalized vector $|v⟩$ is itself an operator, built from the outer product:

Apply $P_v$ to $|\psi ⟩$ and you get the component of the state that lies along $|v⟩$. Three different objects then follow, and keeping them straight is most of this chapter:

• The amplitude $⟨v\mid \psi ⟩$ — a complex number, the overlap itself.
• The probability ${\left|⟨v\mid \psi ⟩\right|}^2$ — the Born rule; a real number between 0 and 1.
• The post-measurement state — the projected vector $P_v|\psi ⟩$, renormalized to unit length.

Measurement returns the outcome $v$ with the Born-rule probability, and afterward the system is described by the renormalized projection. A probability is a number; the post-measurement state is a vector. Confusing the two is the classic beginner error, and it poisons every calculation downstream of it.

Worked example: a state with an imaginary amplitude

Take:

The overlap with $|0⟩$:

so the probability of outcome $0$ is ${\left|\frac{1}{\sqrt{2}}\right|}^2=\frac{1}{2}$. The overlap with $|1⟩$:

and the probability of outcome $1$ is ${\left|\frac{i}{\sqrt{2}}\right|}^2=\frac{1}{2}$. Watch what happened to the $i$: the squared magnitude erased it. The phase was physically real while the state evolved — a later gate could have turned it into a different distribution — but a probability only ever sees magnitudes. If the outcome is $0$, the post-measurement state is simply $|0⟩$.

Where the intuition fails

The first trap is forgetting the conjugate transpose. Real vectors let you get away with it, which is exactly how the habit forms; the first complex vector you meet punishes it. The second trap is treating the probability and the post-measurement state as one object. They answer different questions: how often, versus what remains.

The third trap is overstating distinguishability. If two states are not orthogonal, no measurement — on one copy, in any basis, with any cleverness — identifies them perfectly. A claim that quietly assumes perfect discrimination of non-orthogonal states is a claim that violates the Born rule.

The engineering view

For a computer scientist, an inner product is a similarity query with strict physical semantics. Overlaps appear everywhere: state-preparation checks, amplitude amplification, quantum kernels, error-correction syndromes, and verification protocols all reduce, at some point, to computing $⟨a\mid b⟩$ and deciding what it means.

What it means depends on the task. In an algorithm an overlap may be a success amplitude; in verification it may compare a prepared state against a target; in communication it may bound how well two signals can be told apart. The evidence type differs too: a simulator can report state fidelity directly, while hardware gives you sampled outcomes. Related, but not interchangeable — a test plan that mixes them will lie to you.

What this buys you in diligence

When a paper or pitch claims to distinguish states, classify data, or verify outputs, ask which inner product, projector, or measurement does the work. A claim that leaps from a nonzero overlap to certainty is hiding either an impossible discrimination or a very weak signal. The arithmetic in this chapter is usually enough to tell which.

Exercise

Project and update. Pick any normalized two-component state with at least one imaginary amplitude — for example $\frac{3}{5}|0⟩+\frac{4i}{5}|1⟩$ — and work it completely by hand.

• Submit: the projection onto $|0⟩$ and onto $|1⟩$, both Born-rule probabilities, and the normalized post-measurement state for each outcome.
• Check: confirm at least one overlap used the conjugate transpose, build one projector as an outer product, and verify the two probabilities sum to $1$.
• Repair: if the amplitude–probability–state chain still blurs, redo Chapter 9 (Vectors, Bases, and Amplitudes) before going on.

Check your understanding

Answer without notes: why can two non-orthogonal states never be distinguished perfectly in a single shot?

A passing answer explains that any measurement basis comes with its own projectors, that the Born rule assigns probabilities through overlaps with those projectors, and that a nonzero overlap between the two states forces at least one outcome to be consistent with both. It does not appeal to imperfect hardware.

Oral defense: show that Born-rule probabilities over a complete orthonormal basis always sum to one, using only the projector definition.

If you get stuck

If conjugation, overlap, probabilities, and state updates feel like four unrelated facts rather than one chain, work back through Chapter 9 (Vectors, Bases, and Amplitudes) and then return here. The chain is the chapter.