Composite Systems & Tensor Products
Two qubits, four amplitudes: how state spaces multiply.
Put two qubits side by side and you do not get numbers, you get . The tensor product is the rule by which quantum state spaces multiply, and everything about entanglement, from Bell states to the exponential cost of classical simulation, follows from it.
What you’ll learn
- Compute tensor (Kronecker) products of vectors and matrices, and use the mixed-product rule.
- Write the matrices of local gates such as and of CNOT, and prove that CNOT is not a product of single-qubit operators.
- Decide whether a two-qubit pure state is a product state with the determinant test .
- Compute outcome probabilities and post-measurement states when only one subsystem is measured, and the reduced density matrix .
The composite-system postulate
If system A has state space and system B has state space , the composite system AB lives in the tensor product . If is an orthonormal basis of and one of , then the products form an orthonormal basis of , so
Shorthand: .
In coordinates the tensor product is the Kronecker product: each entry of the first vector multiplies a full copy of the second,
The basis order reads each label as a binary number with qubit 0 as the leftmost, most significant bit. This course and most physics texts use this convention; some software, notably Qiskit, uses the reverse.
Think first Why the tensor product, and not simply stacking the two vectors on top of each other (the direct sum )?
The direct sum describes a system that is either A or B, like one particle in one of two boxes; dimensions add. Two systems that both exist need an amplitude for every combination of basis states, and independent preparations must give multiplying probabilities , which bilinearity guarantees. For two qubits both constructions happen to have dimension 4 (), but for qubits the direct sum has dimension and the tensor product .
What is the dimension of the joint state space of three qubits and one qutrit ()?
Show the answer
C.
Dimensions multiply: .
Operators on composite systems
Operators combine the same way. For acting on and on , the operator is defined by
and extended linearly. Its matrix is the Kronecker product, a block matrix whose blocks are :
The workhorse rule is the mixed-product property . Together with , it shows that tensor products of unitaries are unitary.
A gate on a single qubit is a tensor product with the identity. In our ordering,
The CNOT with qubit 0 as control is a sum of products, "if qubit 0 is do nothing, if it is apply to qubit 1":
Suppose for some single-qubit operators and . Compare both sides on and :
Since would be invertible, , and are all nonzero, and the two equations read
Nonzero product vectors are equal only if their factors are proportional. Comparing the second factors:
No nonzero vector is proportional to both and . So CNOT is a genuinely two-qubit operator. That is exactly why it can create entanglement, while any maps product states to product states.
import numpy as np
ket0 = np.array([1, 0])
I2 = np.eye(2)
X = np.array([[0, 1], [1, 0]])
H = np.array([[1, 1], [1, -1]]) / np.sqrt(2)
P0, P1 = np.diag([1, 0]), np.diag([0, 1]) # |0><0| and |1><1|
CNOT = np.kron(P0, I2) + np.kron(P1, X) # control = qubit 0 (left factor)
psi = CNOT @ np.kron(H, I2) @ np.kron(ket0, ket0)
print(np.round(psi, 3)) # expect [0.707 0. 0. 0.707], i.e. (|00> + |11>)/sqrt(2)The Bell-pair circuit with explicit Kronecker products. np.kron(A, B) puts A on qubit 0, the left factor.
Product or entangled?
Collect the amplitudes of a two-qubit state into its coefficient matrix :
The state is a product state if for some single-qubit states, and entangled otherwise.
A product state has an outer-product coefficient matrix, which has rank one:
So every product state has a vanishing determinant:
Conversely, if , the nonzero matrix has rank one, so for some vectors , . Reading off the entries gives .
For , and : entangled, as the Ground State lessons argued by hand.
Local gates cannot change the verdict. Applying maps the coefficient matrix to (expand to check), so , and for unitaries. The invariant is called the concurrence: for product states and for maximally entangled ones, such as every Bell state.
Which state is entangled?
Show the answer
C.
, so the concurrence is : maximally entangled. It is , and also the output of a CZ gate on . The other three have : they are , and .
Measuring part of a system
Measure only qubit 0 of . The projector for outcome is , and the Born rule and the collapse rule give:
The probability of outcome sums over everything qubit 1 could be:
The state collapses onto the branch where qubit 0 equals ; qubit 1 is left in the normalized row of :
Example: . Outcome has probability and leaves qubit 1 in ; outcome has probability and leaves it in .
What if you never look at qubit 1 and only want to predict measurements on qubit 0? All such predictions are captured by the reduced density matrix , obtained by the partial trace over B. In terms of the coefficient matrix it is simply
For a product state, is pure. For , gives : the maximally mixed state at the center of the Bloch ball.
For , the reduced state is . What is its purity ?
Show the answer
0.78
, so : between (maximally mixed, maximally entangled) and (pure, product). The state is entangled, but not maximally.
Many qubits and the exponential wall
Everything generalizes: qubits live in , with one amplitude per bit string. A product state needs only real parameters (two Bloch angles per qubit), but a general pure state needs . Product states are a tiny corner of the space: almost every state is entangled.
| Qubits | Amplitudes | Memory at 16 bytes per amplitude |
|---|---|---|
| 10 | 16 KiB | |
| 30 | 16 GiB | |
| 50 | 16 PiB | |
| 300 | more bytes than atoms in the observable universe () |
The gap between and is both why quantum computers might be powerful and why simulating them is hard. But it is entanglement that makes the wall real: if every cut through a many-qubit state carries little entanglement, the state can be stored with far fewer numbers. That is the idea behind tensor networks.
Key takeaways
- Composite systems live in : dimensions multiply, and in coordinates is the Kronecker product.
- . A gate on one qubit is ; CNOT is a sum of products and cannot be written as .
- A two-qubit pure state is a product state iff , and local unitaries preserve .
- Measuring one qubit leaves the other in a normalized row of ; ignoring a qubit leaves the reduced state , which is mixed exactly when the pure state is entangled.
Check yourself
In the interactive lesson these are scored with your confidence, and what you miss comes back for review.
What is the amplitude of in ?
Show the answer
D.
. The term is .
For which is a product state?
- only (mod )
- only
- Every : phases never entangle
- No
Show the answer
A. only (mod )
, which vanishes only at . At , and the state is maximally entangled: it is what a controlled- gate makes from .
You measure qubit 1 (the right qubit) of and get . What state is qubit 0 left in?
Show the answer
C.
Keep the terms with qubit 1 equal to : . Qubit 0 is , which is up to a global phase.
What is the reduced state of the singlet ?
- It is undefined, because the state is entangled
Show the answer
C.
and . Like every Bell state, each half is maximally mixed.
Go deeper
- Kronecker Products & Ordering Conventions · Endianness, qubit ordering and getting the matrices right.
- The Partial Trace, Rigorously · Reduced states, and the only consistent way to ignore a subsystem.
- Tensor Networks & Matrix Product States · When entanglement is low, the exponential wall disappears.
Further reading
- M. A. Nielsen & I. L. Chuang, Quantum Computation and Quantum Information, §2.1.7 (tensor products), §2.2.8 (composite systems) and §2.4.3 (the reduced density operator) (Cambridge, 2010).
- N. D. Mermin, Quantum Computer Science: An Introduction, ch. 1 (Cambridge, 2007).
- J. Preskill, Lecture Notes for Physics 219: Quantum Computation (Caltech), chapters on foundations and entanglement.