Skip to content

Tensor Product Representations

A tensor product representation is the natural representation carried by a tensor-product vector space when symmetries act on the factors. It is the representation-theory mechanism behind combining subsystems, adding angular momenta, decomposing singlet and triplet sectors, and organizing coupled versus uncoupled bases.

The linear-algebra construction is Tensor Products. This page explains what happens when the tensor-product factors already carry group or Lie-algebra representations.

Composite quantum systems have tensor-product Hilbert spaces. If each subsystem carries a symmetry representation, the composite system carries a representation too.

For rotations, this is the origin of angular momentum addition. Two spins may be described first by separate labels j1,m1j_1,m_1 and j2,m2j_2,m_2, but the same tensor-product space can be decomposed into total-angular-momentum sectors labeled by J,MJ,M.

The key structural idea is:

tensor product of spaces⟹representation that may decompose as a direct sum.\text{tensor product of spaces} \quad\Longrightarrow\quad \text{representation that may decompose as a direct sum}.

So tensor products and direct sums both appear: tensor products combine subsystems or factors, while direct sums describe the irreducible sectors inside the combined representation.

Let

ρ:G→GL⁡(V),σ:H→GL⁡(W)\rho:G\to \operatorname{GL}(V), \qquad \sigma:H\to \operatorname{GL}(W)

be representations. Their external tensor product is a representation of the product group G×HG\times H on V⊗WV\otimes W:

(ρ⊠σ)(g,h)=ρ(g)⊗σ(h).(\rho\boxtimes\sigma)(g,h) = \rho(g)\otimes\sigma(h).

On a product vector,

(ρ⊠σ)(g,h)(v⊗w)=ρ(g)v⊗σ(h)w.(\rho\boxtimes\sigma)(g,h)(v\otimes w) = \rho(g)v\otimes\sigma(h)w.

The group law works because operator tensor products multiply factor by factor:

(ρ(g1)⊗σ(h1))(ρ(g2)⊗σ(h2))=ρ(g1g2)⊗σ(h1h2).\begin{aligned} &(\rho(g_1)\otimes\sigma(h_1)) (\rho(g_2)\otimes\sigma(h_2))\\ &\qquad = \rho(g_1g_2)\otimes\sigma(h_1h_2). \end{aligned}

This is the right structure when two independent symmetry groups act on two factors.

Often the same group acts on both factors. If

ρ1:G→GL⁡(V1),ρ2:G→GL⁡(V2),\rho_1:G\to \operatorname{GL}(V_1), \qquad \rho_2:G\to \operatorname{GL}(V_2),

then the tensor product representation of GG on V1⊗V2V_1\otimes V_2 is

(ρ1⊗ρ2)(g)=ρ1(g)⊗ρ2(g).(\rho_1\otimes\rho_2)(g) = \rho_1(g)\otimes\rho_2(g).

This is the diagonal action: one group element gg is applied to both factors at once.

For quantum rotations, this is the physically relevant action when a single spatial rotation acts on every subsystem. If U1(R)U_1(R) and U2(R)U_2(R) are the rotation representations on two factors, the composite rotation is

U(R)=U1(R)⊗U2(R).U(R) = U_1(R)\otimes U_2(R).

This formula should not be confused with an arbitrary local operation U1⊗U2U_1\otimes U_2 where the two unitary operators can be chosen independently. A symmetry rotation uses the same rotation parameter on both factors.

For a Lie group representation, differentiating the diagonal tensor product gives the Lie-algebra action.

Let X∈gX\in\mathfrak g. If dρ1(X)d\rho_1(X) acts on V1V_1 and dρ2(X)d\rho_2(X) acts on V2V_2, then on V1⊗V2V_1\otimes V_2:

d(ρ1⊗ρ2)(X)=dρ1(X)⊗I+I⊗dρ2(X).d(\rho_1\otimes\rho_2)(X) = d\rho_1(X)\otimes I + I\otimes d\rho_2(X).

This is the algebraic reason total generators are sums. For two angular momenta,

J=J1+J2,\mathbf J = \mathbf J_1+\mathbf J_2,

meaning componentwise

Ji=J1i⊗I+I⊗J2i.J_i = J_{1i}\otimes I + I\otimes J_{2i}.

The identity factors are part of the statement. They specify which tensor factor each operator acts on.

A tensor product of irreducible representations is usually not irreducible. For compact groups and finite-dimensional unitary representations, it decomposes as a direct sum of irreducible pieces:

Va⊗Vb≅⨁cNab  cVc.V_a\otimes V_b \cong \bigoplus_c N_{ab}^{\ \ c}V_c.

Here Va,Vb,VcV_a,V_b,V_c denote irreducible representation spaces, and the nonnegative integers Nab  cN_{ab}^{\ \ c} are multiplicities. They count how many times each irreducible representation appears.

This is representation decomposition, not factorization of states. The left side is a tensor-product space. The right side is the same vector space reorganized as a direct sum of symmetry sectors.

For angular momentum, the relevant compact group is SU(2)SU(2), and the decomposition is especially simple:

Vj1⊗Vj2≅⨁J=∣j1−j2∣j1+j2VJ.V_{j_1}\otimes V_{j_2} \cong \bigoplus_{J=\lvert j_1-j_2\rvert}^{j_1+j_2} V_J.

Each allowed JJ appears once. Dimension counting checks the formula:

(2j1+1)(2j2+1)=∑J=∣j1−j2∣j1+j2(2J+1).(2j_1+1)(2j_2+1) = \sum_{J=\lvert j_1-j_2\rvert}^{j_1+j_2}(2J+1).

The change-of-basis coefficients between the uncoupled product basis and the coupled direct-sum basis are Clebsch–Gordan Coefficients. The angular-momentum derivation is developed in the Symmetry, Angular Momentum, and Spin treatment.

For two angular momenta, the uncoupled basis is

∣j1,m1⟩⊗∣j2,m2⟩.\lvert j_1,m_1\rangle\otimes\lvert j_2,m_2\rangle.

It diagonalizes J12,J1z,J22,J2zJ_1^2,J_{1z},J_2^2,J_{2z}.

The coupled basis is

∣j1,j2;J,M⟩.\lvert j_1,j_2;J,M\rangle.

It diagonalizes J12,J22,J2,JzJ_1^2,J_2^2,J^2,J_z, where

J=J1+J2.\mathbf J=\mathbf J_1+\mathbf J_2.

Both bases describe the same tensor-product space. The difference is which commuting observables have been diagonalized. The Clebsch–Gordan coefficients are the entries of the unitary change-of-basis matrix between them.

Let V1/2≅C2V_{1/2}\cong\mathbb C^2 be the spin-1/21/2 representation space. The tensor product has dimension four:

dim⁡(V1/2⊗V1/2)=2⋅2=4.\dim(V_{1/2}\otimes V_{1/2}) = 2\cdot2 = 4.

The SU(2)SU(2) decomposition is

V1/2⊗V1/2≅V1⊕V0.V_{1/2}\otimes V_{1/2} \cong V_1\oplus V_0.

The dimensions match:

4=3+1.4 = 3+1.

The three-dimensional sector V1V_1 is the spin-11 triplet. The one-dimensional sector V0V_0 is the spin-00 singlet. A representative singlet vector is

∣0,0⟩=12(∣↑↓⟩−∣↓↑⟩).\lvert 0,0\rangle = \frac{1}{\sqrt2} \left( \lvert\uparrow\downarrow\rangle - \lvert\downarrow\uparrow\rangle \right).

The full coupled basis and its physical interpretation are developed in Two Spin-1/2 Particles.

Tensor product representations organize the space on which a symmetry acts. They do not say that every state is a product state.

The tensor-product space V1⊗V2V_1\otimes V_2 contains product vectors such as v1⊗v2v_1\otimes v_2, but it also contains linear combinations that cannot be factored. In quantum mechanics, those are entangled states when the tensor factors are physical subsystems.

The direct-sum decomposition into irreducible sectors is a different structure. For example, the singlet/triplet decomposition reorganizes the two-spin tensor-product space by total spin. The singlet state is still a vector in the original tensor-product Hilbert space.

  • Confusing the product group action G×HG\times H with the diagonal action of one group GG on both factors.
  • Forgetting identity factors in total generators such as J1i⊗I+I⊗J2iJ_{1i}\otimes I+I\otimes J_{2i}.
  • Treating j1⊗j2=J1⊕J2⊕⋯j_1\otimes j_2=J_1\oplus J_2\oplus\cdots as arithmetic rather than representation decomposition.
  • Assuming a tensor product of irreducible representations is automatically irreducible.
  • Confusing a direct-sum decomposition into symmetry sectors with a tensor-product subsystem split.
  • Treating Clebsch–Gordan coefficients as dynamics instead of change-of-basis coefficients.
  • Forgetting that phase conventions affect the signs of coupled basis vectors.
  • W. Fulton and J. Harris, Representation Theory: A First Course, Springer, 1991.
  • B. C. Hall, Lie Groups, Lie Algebras, and Representations, 2nd ed., Springer, 2015.
  • J. F. Cornwell, Group Theory in Physics, Vol. 1, Academic Press, 1984.
  • A. R. Edmonds, Angular Momentum in Quantum Mechanics, Princeton University Press, 1957.
  • D. A. Varshalovich, A. N. Moskalev, and V. K. Khersonskii, Quantum Theory of Angular Momentum, World Scientific, 1988.
  • J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020.
  1. Verify that the external tensor product is a representation of G×HG\times H.
Solution

Compute

(ρ⊠σ)(g1,h1)(ρ⊠σ)(g2,h2)=(ρ(g1)⊗σ(h1))(ρ(g2)⊗σ(h2))=ρ(g1g2)⊗σ(h1h2)=(ρ⊠σ)(g1g2,h1h2).\begin{aligned} &(\rho\boxtimes\sigma)(g_1,h_1) (\rho\boxtimes\sigma)(g_2,h_2)\\ &\qquad = (\rho(g_1)\otimes\sigma(h_1)) (\rho(g_2)\otimes\sigma(h_2))\\ &\qquad = \rho(g_1g_2)\otimes\sigma(h_1h_2)\\ &\qquad = (\rho\boxtimes\sigma)(g_1g_2,h_1h_2). \end{aligned}

The identity element maps to IV⊗IWI_V\otimes I_W, so the representation axioms hold.

  1. Derive the Lie-algebra generator rule for a diagonal tensor product.
Solution

For a one-parameter subgroup g(t)=exp⁡(tX)g(t)=\exp(tX),

(ρ1⊗ρ2)(g(t))=ρ1(g(t))⊗ρ2(g(t)).(\rho_1\otimes\rho_2)(g(t)) = \rho_1(g(t))\otimes\rho_2(g(t)).

Differentiate at t=0t=0 and use the product rule:

ddt∣t=0ρ1(g(t))⊗ρ2(g(t))=dρ1(X)⊗I+I⊗dρ2(X).\left.\frac{d}{dt}\right|_{t=0} \rho_1(g(t))\otimes\rho_2(g(t)) = d\rho_1(X)\otimes I + I\otimes d\rho_2(X).
  1. Check the dimension identity for j1=1j_1=1 and j2=1/2j_2=1/2.
Solution

The tensor-product dimension is

(2j1+1)(2j2+1)=3⋅2=6.(2j_1+1)(2j_2+1) = 3\cdot2 = 6.

The allowed total angular momenta are

J=12,32.J=\frac12,\frac32.

The sum of sector dimensions is

(2⋅12+1)+(2⋅32+1)=2+4=6.(2\cdot\tfrac12+1) + (2\cdot\tfrac32+1) = 2+4 = 6.
  1. Explain why V1/2⊗V1/2≅V1⊕V0V_{1/2}\otimes V_{1/2}\cong V_1\oplus V_0 is not an equation of ordinary numbers.
Solution

The symbols V1/2V_{1/2}, V1V_1, and V0V_0 denote representation spaces, not numbers. The left side is the four-dimensional tensor-product representation of two spin-1/21/2 spaces. The right side is the same representation decomposed into a three-dimensional spin-11 irreducible sector and a one-dimensional spin-00 irreducible sector.

  1. In the diagonal rotation action on two spin spaces, why is the total generator Ji=J1i⊗I+I⊗J2iJ_i=J_{1i}\otimes I+I\otimes J_{2i} rather than J1i⊗J2iJ_{1i}\otimes J_{2i}?
Solution

A small rotation acts as

U1(t)⊗U2(t)=(I−iℏtJ1i+O(t2))⊗(I−iℏtJ2i+O(t2)).U_1(t)\otimes U_2(t) = \left(I-\frac{i}{\hbar}tJ_{1i}+O(t^2)\right) \otimes \left(I-\frac{i}{\hbar}tJ_{2i}+O(t^2)\right).

Keeping only first-order terms gives

I⊗I−iℏt(J1i⊗I+I⊗J2i)+O(t2).I\otimes I - \frac{i}{\hbar}t \left( J_{1i}\otimes I+I\otimes J_{2i} \right) + O(t^2).

The product J1i⊗J2iJ_{1i}\otimes J_{2i} appears only at order t2t^2, not as the infinitesimal generator.