Direct Sums versus Tensor Products

Direct sums and tensor products answer different structural questions.

A direct sum collects alternatives, sectors, or invariant subspaces into one larger Hilbert space. A tensor product describes simultaneous degrees of freedom or subsystems that exist together. Confusing the two changes dimensions, basis labels, operator structure, and the meaning of entanglement.

The quickest finite-dimensional test is

$$\dim(\mathcal H_A\oplus\mathcal H_B) = \dim\mathcal H_A+\dim\mathcal H_B, \qquad \dim(\mathcal H_A\otimes\mathcal H_B) = \dim\mathcal H_A\,\dim\mathcal H_B.$$

If one system has two levels and another has three levels, the composite system has six basis states, not five. The five-dimensional direct sum would describe a space with a two-dimensional sector and a three-dimensional sector, not two subsystems present at the same time.

Helpful background. Product Bases explains paired subsystem labels; elementary vector spaces and inner products are sufficient for the direct-sum comparison.

Direct Sum as Alternatives or Sectors

For two Hilbert spaces $\mathcal H_1$ and $\mathcal H_2$, the direct sum

$$\mathcal H = \mathcal H_1\oplus\mathcal H_2$$

consists of ordered pairs

$$\psi_1\oplus\psi_2, \qquad \psi_1\in\mathcal H_1,\quad \psi_2\in\mathcal H_2,$$

with inner product

$$\langle\psi\vert\phi\rangle_{\mathcal H} = \langle\psi_1\vert\phi_1\rangle_{\mathcal H_1} + \langle\psi_2\vert\phi_2\rangle_{\mathcal H_2}.$$

The two summands are orthogonal subspaces of one Hilbert space. A state may lie entirely in one sector, such as

$$\psi_1\oplus0,$$

or it may be a coherent superposition with components in several sectors,

$$\psi_1\oplus\psi_2.$$

The word “alternative” should therefore be read carefully. Direct sums do not forbid superpositions across sectors by themselves. They say that the Hilbert space has sector labels rather than independent subsystem slots. Whether coherent relative phases between sectors are physically accessible depends on the allowed observables, dynamics, and reference frames.

An operator that preserves the sectors is block diagonal:

$$O = \begin{pmatrix} O_1&0\\ 0&O_2 \end{pmatrix}.$$

An operator with off-diagonal blocks maps one sector into another:

$$O = \begin{pmatrix} O_{11}&O_{12}\\ O_{21}&O_{22} \end{pmatrix}.$$

Whether such sector-changing blocks are allowed is a physical question, not a definition of direct sum.

Tensor Product as Simultaneous Composition

For two subsystems $A$ and $B$, the Hilbert space is

$$\mathcal H_{AB} = \mathcal H_A\otimes\mathcal H_B.$$

A product vector has the form

$$\lvert\psi\rangle_A\otimes\lvert\phi\rangle_B.$$

This means that both subsystem slots are present. It is not a choice between an $A$ state and a $B$ state. If $\{\lvert i\rangle_A\}$ and $\{\lvert j\rangle_B\}$ are bases, the product basis is

$$\lvert i\rangle_A\otimes\lvert j\rangle_B,$$

one basis vector for each ordered pair of labels.

The operator structure is also different. A local operator on $A$ is embedded as

$$O_A\otimes I_B,$$

while a local operator on $B$ is

$$I_A\otimes O_B.$$

This local-operator structure is what supports reduced states, local measurements, correlations, and entanglement. A direct sum has blocks; a tensor product has factors.

Comparison Table

Question	Direct sum	Tensor product
Symbol	$\mathcal H_1\oplus\mathcal H_2$	$\mathcal H_A\otimes\mathcal H_B$
Basic meaning	sectors or alternatives in one Hilbert space	simultaneous subsystem factors
Finite dimension	$m+n$	$mn$
Basis labels	one label from one sector	ordered pairs of subsystem labels
Typical operator form	block matrices	sums of product operators
Entanglement?	not defined between summands as ordinary subsystem entanglement	defined relative to a tensor-product split

The last row is often the most important. Entanglement is defined relative to a specified tensor-factor decomposition or an explicitly specified algebraic substitute. Which local operations are experimentally available is a further physical question. A direct sum by itself gives orthogonal sectors, not two parties.

Spin and Spatial Degrees of Freedom

For one spin-$1/2$ particle moving in space, the Hilbert space is

$$\mathcal H = L^2(\mathbb R^3)\otimes\mathbb C^2.$$

The particle has both a spatial wavefunction and a spin degree of freedom. A simple product state has the form

$$\Psi(\mathbf x) = \psi(\mathbf x)\chi,$$

where $\chi\in\mathbb C^2$ is a spinor. More general spinor wavefunctions need not factor:

$$\Psi(\mathbf x) = \psi_\uparrow(\mathbf x)\lvert\uparrow\rangle + \psi_\downarrow(\mathbf x)\lvert\downarrow\rangle.$$

After choosing the $\uparrow,\downarrow$ spin basis, there is an isomorphism

$$L^2(\mathbb R^3)\otimes\mathbb C^2 \cong L^2(\mathbb R^3)\oplus L^2(\mathbb R^3),$$

because a spinor wavefunction can be represented by two spatial components. This useful representation does not mean that spin and space were combined by an either-or direct sum. The underlying degrees of freedom are simultaneous; the direct-sum display appears after decomposing the spin factor into basis lines.

This distinction becomes important for spin-orbit coupling. An operator such as $\mathbf L\cdot\mathbf S$ acts jointly on spatial and spin degrees of freedom. It belongs naturally to the tensor-product description.

Representation Decompositions

Direct sums also appear when a tensor-product representation decomposes into irreducible pieces. For two spin-$1/2$ systems,

$$\frac12\otimes\frac12 = 1\oplus0$$

is representation shorthand. It means that the four-dimensional two-spin Hilbert space decomposes into a three-dimensional triplet subspace and a one-dimensional singlet subspace:

$$\mathbb C^2\otimes\mathbb C^2 \cong \mathcal H_{\mathrm{triplet}} \oplus \mathcal H_{\mathrm{singlet}}.$$

The physical system is still two spins composed by a tensor product. The direct sum appears after reorganizing the tensor-product space according to total angular momentum.

The same pattern occurs throughout quantum mechanics: first construct the composite Hilbert space, then decompose it into symmetry sectors when a symmetry makes that useful.

Superselection Sectors

A superselection rule is often expressed through a direct-sum decomposition into sectors:

$$\mathcal H = \bigoplus_q \mathcal H_q,$$

where $q$ may label a superselected charge. Conservation of a quantity alone is not sufficient: a Hamiltonian can conserve it while other accessible measurements detect superpositions of its eigenstates. The extra restriction is that allowed observables are block diagonal in q; allowed dynamics must also respect the corresponding operational restriction.

If $\Pi_q$ projects onto $\mathcal H_q$, a sector-preserving observable satisfies

$$A = \sum_q \Pi_q A\Pi_q.$$

Then the off-diagonal coherences of a density operator,

$$\Pi_q\rho\Pi_{q'}, \qquad q\ne q',$$

do not affect expectation values of such observables. Equivalently, $\rho$ and $\sum_q\Pi_q\rho\Pi_q$ give the same allowed measurement statistics. The broader symmetry-side language is summarized in Superselection Sectors Preview, while particle-number details are treated in the Particle-Number Superselection Preview.

Do not infer superselection from the symbol $\oplus$ alone. A direct sum is a Hilbert-space construction. A superselection rule is an additional statement about allowed operations and observables.

Fock Space as a Direct Sum of Particle-Number Sectors

Fock space is the standard place where both constructions appear together. For a one-particle Hilbert space $\mathcal h$, the bosonic Fock space is

$$\mathcal F_B(\mathcal h) = \bigoplus_{N=0}^{\infty} \operatorname{Sym}^N\mathcal h.$$

The fermionic Fock space is

$$\mathcal F_F(\mathcal h) = \bigoplus_{N=0}^{\infty} \wedge^N\mathcal h.$$

Inside each fixed-$N$ sector, the construction comes from tensor powers of the one-particle Hilbert space:

$$\mathcal h^{\otimes N}.$$

For identical bosons one takes the symmetric subspace $\operatorname{Sym}^N\mathcal h$. For identical fermions one takes the antisymmetric subspace $\wedge^N\mathcal h$. The vacuum sector at $N=0$ is $\mathbb C$. Fock space forms the Hilbert direct sum over N: its vectors have components $\Psi_N$ satisfying $\sum_N\|\Psi_N\|^2<\infty$, with normalization requiring that sum to equal one.

Thus Fock space is not a tensor product over particle-number sectors. A vector such as

$$\alpha\lvert0\rangle + \beta\lvert1_f\rangle$$

has a vacuum component and a one-particle component in a direct sum. It is not entanglement between a “vacuum subsystem” and a “one-particle subsystem.” Whether the relative phase between those components is operationally meaningful depends on the physical setting.

A superposition between summands

The two-dimensional space $\mathbb C\oplus\mathbb C$ can describe one qubit with basis $|0\rangle,|1\rangle$. Its vector $|+\rangle=(|0\rangle+|1\rangle)/\sqrt2$ has coherence between the summands. The off-diagonal observable

$$X=|0\rangle\langle1|+|1\rangle\langle0|$$

detects this coherence: $\langle X\rangle=1$, whereas the equal diagonal mixture has $\langle X\rangle=0$. Restricting all observables to diagonal matrices would remove this distinction. The restriction, not the direct-sum notation, is what makes the phase inaccessible.

By contrast, $\mathbb C^2\otimes\mathbb C^2$ describes two qubits with four basis vectors and supports entanglement between them. The dimension and the local operator structure together identify the physical interpretation.

Common pitfalls

A direct sum does not prohibit coherence. Off-diagonal operators can compare different summands unless an additional physical restriction forbids them.

Equal dimensions do not identify subsystem structure. Two qubits and a single four-level system both use a four-dimensional Hilbert space. A specified pair of local operator algebras is what gives the two-qubit interpretation.

Symmetry sectors are not separate particles. Singlet and triplet subspaces decompose the same two-spin system; particle-number sectors decompose the same Fock space.

References

• S. D. Bartlett, T. Rudolph, and R. W. Spekkens, “Reference frames, superselection rules, and quantum information,” Reviews of Modern Physics 79, 555–609, 2007. arXiv.
• A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems, McGraw-Hill, 1971.
• B. C. Hall, Quantum Theory for Mathematicians, Springer, 2013. Publisher.
• M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information, 10th anniversary ed., Cambridge University Press, 2010.

Exercises

1. Let $\dim\mathcal H_A=2$ and $\dim\mathcal H_B=3$. What are the dimensions of $\mathcal H_A\oplus\mathcal H_B$ and $\mathcal H_A\otimes\mathcal H_B$? Which one describes two subsystems present together?

Solution

The direct sum has dimension

$$2+3=5.$$

The tensor product has dimension

$$2\cdot3=6.$$

The tensor product describes two subsystems present together. Its basis vectors are labeled by ordered pairs, one label from $\mathcal H_A$ and one from $\mathcal H_B$.

2. A spin-$1/2$ particle moving on a line has Hilbert space $L^2(\mathbb R)\otimes\mathbb C^2$. Explain why this is sometimes displayed as $L^2(\mathbb R)\oplus L^2(\mathbb R)$, and why that display should not be confused with the basic physical composition rule.

Solution

After choosing a spin basis $\lvert\uparrow\rangle,\lvert\downarrow\rangle$, a spinor wavefunction can be written as two spatial components:

$$\Psi(x) = \psi_\uparrow(x)\lvert\uparrow\rangle + \psi_\downarrow(x)\lvert\downarrow\rangle.$$

This identifies the state with a pair of functions, so

$$L^2(\mathbb R)\otimes\mathbb C^2 \cong L^2(\mathbb R)\oplus L^2(\mathbb R).$$

The physical degrees of freedom are still simultaneous position and spin degrees of freedom. The direct-sum display is a component representation chosen after selecting a spin basis.

3. In the formula $\frac12\otimes\frac12=1\oplus0$, what role is played by the tensor product and what role is played by the direct sum?

Solution

The tensor product constructs the Hilbert space of two spin-$1/2$ systems:

$$\mathbb C^2\otimes\mathbb C^2.$$

The direct sum describes how that four-dimensional space decomposes into irreducible total-angular-momentum sectors:

$$\mathcal H_{\mathrm{triplet}} \oplus \mathcal H_{\mathrm{singlet}}.$$

The triplet sector has dimension three and the singlet sector has dimension one.

4. Why is the bosonic Fock space $\mathcal F_B(\mathcal h)=\bigoplus_N\operatorname{Sym}^N\mathcal h$ not a tensor product over $N$?

Solution

Each value of $N$ labels a fixed-particle-number sector. A Fock-space vector has components in these sectors, and its squared norm is the sum $\sum_N\|\Psi_N\|^2$ of the squared sector norms. The sectors are alternatives in one Hilbert space, possibly coherently superposed if the physical setting permits.

A tensor product over $N$ would describe all fixed-number sectors as simultaneous independent subsystems. That is not what particle number means. The tensor products occur inside the fixed-$N$ sectors through $\mathcal h^{\otimes N}$, followed by symmetrization for bosons.