Quantum Foundations for Finance

Every quantum-finance pitch leans on four words — qubit, superposition, entanglement, gate — and a promise that they add up to exponential free lunch. They do not. But you cannot audit a “quantum advantage in option pricing” claim if those four words are fog, so this lesson turns them into working intuition. Not the physicist’s version with bra-ket gymnastics and Hilbert-space measure theory — the quant’s version: just enough machinery to know what a quantum computer can and cannot do, and which of the two very different machines you are even talking about.

Here’s the thesis up front: a quantum computer is not a magic parallel processor that tries all answers at once and hands you the best one. It is a device that prepares a delicate cloud of possibilities, then — through carefully arranged interference — makes the answer you want loud and everything else quiet, so that a single measurement is likely to return it. The whole game is amplitude bookkeeping. Get that one idea and the rest of the course (the quadratic Monte-Carlo speedup, QAOA, the data-loading wall) snaps into place.

The qubit vs the bit

Analogy. A bit is a light switch: up or down, on or off, end of story. A qubit is more like a spinning coin before it lands. While it spins it is genuinely neither heads nor tails — it carries a propensity toward each — and the instant it hits the table it commits to exactly one face. The spin is the superposition; the landing is the measurement.

Definition. A qubit’s state is a superposition of the two basis states |0⟩ and |1⟩:

$|\psi\rangle = a\,|0\rangle + b\,|1\rangle, \qquad |a|^2 + |b|^2 = 1.$

The numbers $a$ and $b$ are amplitudes (complex numbers, in general). They are not probabilities — you square their magnitudes to get probabilities. When you measure the qubit in the |0⟩/|1⟩ basis, you get outcome 0 with probability $|a|^2$ and outcome 1 with probability $|b|^2$, and the state instantly collapses to whichever you saw. The normalization $|a|^2 + |b|^2 = 1$ just says “the probabilities of the two outcomes sum to 1.”

Worked example. Take the most famous single-qubit state, the equal superposition:

$|\psi\rangle = \tfrac{1}{\sqrt{2}}\,|0\rangle + \tfrac{1}{\sqrt{2}}\,|1\rangle, \qquad a = b = \tfrac{1}{\sqrt{2}}.$

Check the normalization first: $|a|^2 + |b|^2 = \left(\tfrac{1}{\sqrt 2}\right)^2 + \left(\tfrac{1}{\sqrt 2}\right)^2 = \tfrac{1}{2} + \tfrac{1}{2} = 1.$ Good. Now the measurement probabilities: $P(0) = |a|^2 = \tfrac{1}{2}$ and $P(1) = |b|^2 = \tfrac{1}{2}$ — a perfect 50/50 coin. Notice that the amplitude was $\tfrac{1}{\sqrt 2} \approx 0.707$, not $0.5$; you square it to land at the 50% you expected. Get an unequal state, say $a = \tfrac{\sqrt 3}{2}$ and $b = \tfrac{1}{2}$, and you get $P(0) = \tfrac{3}{4} = 75\%$ and $P(1) = \tfrac{1}{4} = 25\%$ — again the squares, and again they sum to 1.

Why it matters for finance

Amplitudes can be negative (or complex), and probabilities cannot — that sign is the entire source of quantum advantage. Two paths to the same outcome with amplitudes $+0.5$ and $-0.5$ cancel (destructive interference) to probability zero; classical probabilities, always positive, can only pile up. Every speedup later in this course — the $1/N$ Monte-Carlo error decay of amplitude estimation especially — is squeezed out of arranging amplitudes so the wrong answers cancel and the right one survives. No negative amplitudes, no advantage.

Superposition and the “2ⁿ states at once” claim

Analogy. Picture a giant orchestra where every possible answer to your problem is a musician playing a note. In superposition, all of them play at once — a deafening chord containing every candidate. But you are allowed exactly one recording, and a microphone captures whichever note is loudest. Quantum computing is not “hear every musician separately.” It is conducting: arranging the music so that destructive interference silences the wrong notes and constructive interference swells the right one, so the single note your microphone grabs is overwhelmingly likely to be the answer.

Definition. With $n$ qubits, the state lives in a space of $2^n$ basis states, and the general superposition is

$|\psi\rangle = \sum_{x=0}^{2^n - 1} c_x\, |x\rangle, \qquad \sum_{x} |c_x|^2 = 1,$

where each $|x\rangle$ is one of the $2^n$ possible $n$-bit strings and $c_x$ is its amplitude. Yes, that is exponentially many amplitudes evolving together — the source of the “exponential power” headline. But the measurement rule is unchanged: one measurement returns a single string $x$ with probability $|c_x|^2$, and the rest of the state is gone. You never get to read the whole vector of amplitudes. The “art” of a quantum algorithm is choreographing the amplitudes so that, before you measure, $|c_x|^2$ is near 1 for the $x$ you want and near 0 for the $2^n - 1$ you don’t.

Worked example. Take $n = 3$ qubits, all placed in equal superposition. There are $2^3 = 8$ basis states (000 through 111), each with amplitude $c_x = \tfrac{1}{\sqrt{8}}$. Each one’s measurement probability is $|c_x|^2 = \tfrac{1}{8} = 12.5\%$. So a naive measurement is just a uniform random 3-bit string — worse than useless, you could have flipped three coins. Now suppose a clever circuit (Grover-style) bumps the amplitude of the target string 101 up to, say, $0.97$ while shrinking the other seven. Then $P(101) = 0.97^2 \approx 0.94$: a single measurement returns your answer 94% of the time. Same $2^n$ states held throughout — the only thing that changed is who got the amplitude. That redistribution is the whole job.

Why it matters for finance

This is the single most important corrective for a finance audience, because almost every hype cycle rests on the free-parallelism fallacy. When you reach amplitude estimation in lesson 2, you will see the speedup is a quadratic one — turning Monte Carlo’s $1/\sqrt{N}$ error decay into $1/N$ — precisely because it is interference-limited, not magic. And when optimization vendors claim a quantum box “evaluates every portfolio at once,” you will know to ask: where does the amplitude on the optimal portfolio actually come from? If they can’t answer, the speedup is imaginary.

Entanglement

Analogy. Imagine two coins minted together with a strange rule: each spins independently and lands heads or tails 50/50, yet whenever you check them they always land on opposite faces — one heads, one tails — no matter how far apart you’ve carried them. Neither coin has a predetermined face (each is genuinely 50/50 on its own), yet the pair shares a destiny that no description of either coin alone can capture. That shared, non-factorable destiny is entanglement. There is no hidden note slipped between them and no signal flying across the room — the correlation is baked into the joint state.

Definition. A two-qubit state is entangled when it cannot be factored into a product of two single-qubit states. A factorable (separable) state looks like $|\psi\rangle = |\phi_A\rangle \otimes |\phi_B\rangle$. The canonical entangled state, the Bell state, does not:

$|\Phi^+\rangle = \tfrac{1}{\sqrt{2}}\big(|00\rangle + |11\rangle\big).$

Try to write that as (something for qubit A) times (something for qubit B) and you fail — there is no such pair. The correlation lives in the joint description, not in either qubit.

Worked example. Measure the first qubit of $|\Phi^+\rangle$. The amplitudes on |00⟩ and |11⟩ are each $\tfrac{1}{\sqrt 2}$, so you get outcome 0 with probability $|\tfrac{1}{\sqrt 2}|^2 = \tfrac{1}{2}$ and outcome 1 with probability $\tfrac{1}{2}$ — a fair coin, individually. But the instant you see 0 on the first qubit, the |11⟩ branch is annihilated and the state collapses to |00⟩, so the second qubit is now certainly 0 ($P = 1$). See 1 instead, and the second is certainly 1. Individually random, perfectly correlated jointly: that is the signature you cannot reproduce by handing each qubit its own private probability table.

Why it matters for finance

Entanglement is the resource that makes the state space genuinely $2^n$-dimensional rather than $n$ independent coins. Without it, $n$ qubits would carry no more correlated structure than $n$ separate biased coins, and there would be nothing to compute with. The interference tricks that power amplitude estimation and QAOA all rely on building up entanglement across the register. For the quant, the useful mental hook is correlation with no classical factorization — the same instinct you use when a covariance matrix refuses to diagonalize into independent factors, pushed to a regime classical probability literally cannot represent.

Quantum gates and circuits

Analogy. A classical AND gate is a paper shredder: feed it two bits, get one out, and you can never reconstruct the inputs — information was destroyed. A quantum gate is more like a Rubik’s-cube twist: it rearranges the state in a fully reversible way, and the exact opposite twist always undoes it. A quantum circuit is a choreographed sequence of those twists applied to your qubits, and only at the very end do you “open your eyes” — measure — and read out bits.

Definition. A quantum gate is a unitary operation $U$ on the qubits: unitary means $U^\dagger U = I$, which guarantees it is reversible (its inverse is $U^\dagger$) and preserves the total probability ($\sum |c_x|^2 = 1$ stays true). Familiar gates:

• The Hadamard ($H$) turns |0⟩ into the equal superposition $\tfrac{1}{\sqrt 2}(|0\rangle + |1\rangle)$ — it creates superposition from a definite state.
• A rotation gate nudges the amplitudes by a chosen angle, used to load numbers and angles into the state.
• The CNOT (controlled-NOT) flips a target qubit only when the control qubit is 1 — a two-qubit gate that creates entanglement.

A circuit is a sequence of gates applied to an initialized register, ending in one measurement. Two numbers describe a circuit’s size: its width (how many qubits) and its depth — the number of sequential gate layers from input to measurement.

Worked example — building a Bell state in two gates. Start with both qubits in |00⟩. Apply $H$ to the first qubit: the state becomes $\tfrac{1}{\sqrt 2}(|00\rangle + |10\rangle)$ — the first qubit is now a fair superposition, the second still 0. Now apply a CNOT with the first qubit as control and the second as target: the |00⟩ term is left alone (control is 0), while the |10⟩ term flips its second qubit to give |11⟩. The result is $\tfrac{1}{\sqrt 2}(|00\rangle + |11\rangle) = |\Phi^+\rangle$ — the entangled Bell state from the last section, built with depth 2 (Hadamard, then CNOT). Two gates; an entangled pair.

When to use it

Think in gates and depth whenever you evaluate a quantum algorithm’s feasibility, not just its theoretical speedup. Two algorithms can promise the same answer; the one with shallower depth and fewer two-qubit gates (the noisy, expensive ones) is the one that might actually run on near-term hardware. When lesson 2’s amplitude estimation and lesson 3’s QAOA show up, you’ll weigh them partly by the circuit depth they demand — because depth, not cleverness, is usually what stands between a quantum result and a press release.

Gate model vs quantum annealing

Analogy. The gate model is a general-purpose computer — like your laptop’s CPU, it runs any algorithm you can express, from option pricing to chess. Quantum annealing is a single-purpose appliance — like a marble rolling down a bumpy landscape until it settles in the lowest valley. The marble is fantastic at one job (finding the lowest point) and useless at everything else. Asking an annealer to run amplitude estimation is like asking a ball bearing to play chess.

Definition. The two hardware paradigms:

• The gate model builds computations from sequences of unitary gates (the circuits above). It is universal: any quantum algorithm — Grover’s search, Shor’s factoring, Quantum Amplitude Estimation (QAE), QAOA — can be expressed as a gate circuit. Hardware: superconducting qubits (IBM, Google) and trapped ions (IonQ, Quantinuum).
• Quantum annealing is special-purpose. It physically prepares a system whose energy encodes your problem, then slowly evolves it so it relaxes into a low-energy state. It solves exactly one shape of problem: minimizing an Ising energy or its equivalent QUBO (Quadratic Unconstrained Binary Optimization), of the form

$\min_{x \in \{0,1\}^n} \; \sum_i h_i\, x_i + \sum_{i<j} J_{ij}\, x_i x_j.$

It cannot run a general circuit. Hardware: D-Wave, with thousands of qubits — but qubits that only anneal, not gate.

Worked example — reading the two spec sheets. Suppose you want option pricing via amplitude estimation. That is a gate algorithm, so a 5,000-qubit annealer scores zero on it — wrong machine entirely; you need the (much smaller) gate-model device. Now suppose you want to pick the best 10-asset subset out of 50 under a cardinality constraint — a discrete QUBO. The annealer’s thousands of qubits become relevant, while the 127-qubit gate machine must emulate the same QUBO through QAOA and may run out of room. The lesson: the right qubit count depends entirely on which paradigm your problem needs, which is why a single “how many qubits?” number tells you almost nothing on its own.

When to use it

Match the machine to the problem shape. If your task is “estimate an expectation or price a derivative,” you want the gate model and amplitude estimation — annealing has nothing to offer. If your task is “choose a discrete subset to minimize a quadratic objective” (asset selection, lot-size constraints), a QUBO formulation opens the door to both annealing and gate-model QAOA, and you compare them — and a classical solver — end to end. The instinct to cultivate: never let a raw qubit count substitute for the question “is this even the right kind of machine?”

Recap

You came in with four fog-words and a promise of free lunch. You leave with working intuition: a qubit is a superposition $a|0\rangle + b|1\rangle$ whose amplitudes square to measurement probabilities; superposition holds $2^n$ amplitudes but yields one sample, so the real game is interference that makes the wanted answer loud; entanglement is correlation no product of independent qubits can fake; gates are reversible unitary twists assembled into circuits whose depth is the clock noise is racing; and the field splits into the universal gate model (QAE, QAOA) versus special-purpose quantum annealing (QUBO minimization on D-Wave). That last split is your compass for the rest of the course.