Many-Body Hilbert Spaces Overview

A many-body Hilbert space is the state space obtained after declaring which degrees of freedom exist, whether particles are distinguishable, which exchange statistics apply, whether particle number is fixed, and which constraints or cutoffs are imposed. It is part of the physical model, not merely a container chosen after the Hamiltonian is written.

The same verbal system can lead to different spaces. “Two particles in four orbitals” could mean two labeled particles in $\mathbb C^4\otimes\mathbb C^4$, two bosons in $\operatorname{Sym}^2\mathbb C^4$, or two fermions in $\bigwedge^2\mathbb C^4$. Those spaces have dimensions $16$, $10$, and $6$, respectively. They encode different physical assumptions before any interaction is specified.

The tensor-product construction, exchange symmetry, and Fock space supply the underlying mathematics. Here we use those constructions to select a working space and distinguish an exact representation change from a physical approximation.

Helpful background. Tensor products describes composition; Fock space and exchange symmetry describe particle sectors.

Degrees of freedom, sectors, and cutoffs

A useful model statement identifies the complete modes, their statistics, the particle numbers or other charges being fixed, and any cutoffs. For a spinful lattice gas, for example, a mode is a site–spin pair. The choices “fixed total number,” “fixed up- and down-spin numbers,” and “all number sectors” lead to different working spaces even with the same Hamiltonian formula.

A finite spatial volume does not by itself make a continuum Hilbert space finite dimensional. A finite-dimensional calculation also needs a mode, energy, or occupation restriction. Conversely, selecting an exactly conserved sector can be an exact reduction for a specified preparation, rather than an approximation.

The One-Particle Hilbert Space

For particle-based problems, the starting object is a one-particle Hilbert space $\mathcal h$. A continuum particle with position, spin, and perhaps an orbital or species label may have

$$\mathcal h = L^2(\Omega,d^dr) \otimes \mathbb C^{g_{\mathrm{spin}}} \otimes \mathbb C^{g_{\mathrm{internal}}}.$$

The factors describe simultaneous one-particle degrees of freedom. A basis vector might be labeled by

$$\lvert\alpha\rangle = \lvert\mathbf r,\sigma,a\rangle,$$

or by momentum, band, and spin labels,

$$\lvert\alpha\rangle = \lvert\mathbf k,n,\sigma\rangle.$$

These are different bases or representations of the same declared one-particle model when connected by a unitary transformation. A finite calculation usually replaces $\mathcal h$ by a retained subspace

$$\mathcal h_M = \operatorname{span} \left\{ \lvert\varphi_1\rangle, \ldots, \lvert\varphi_M\rangle \right\}.$$

The integer $M$ then counts one-particle modes, including all internal labels. A spatial orbital with spin up and the same spatial orbital with spin down are two distinct spin-orbitals.

Species require another declaration

Different species can be represented by separate one-particle spaces. For two distinguishable species $A$ and $B$ with fixed particle numbers,

$$\mathcal H = \mathcal H_{N_A}^{(A)} \otimes \mathcal H_{N_B}^{(B)}.$$

Exchange symmetry is imposed within each identical species, not between physically distinguishable species. If conversion processes change species labels, several species-number sectors are needed. When distinct species are both fermionic, a common ordered Fock construction or graded convention fixes cross-species signs; physical even observables remain compatible with the species decomposition.

Particle tensor products and exchange sectors

For $N$ distinguishable particles sharing an $M$-dimensional one-particle space, $\mathcal h^{\otimes N}$ has dimension $M^N$. Its factors identify independently addressable particles. For identical particles the same formal slot product is restricted to an exchange sector:

Statistics	Fixed-number space	Dimension for M complete modes
Bosons	$\operatorname{Sym}^N\mathcal h$	$\binom{N+M-1}{N}$
Fermions	$\bigwedge^N\mathcal h$	$\binom MN$ for $0\le N\le M$

The fermionic sector is zero when $N>M$. Exchange operators constructs the projectors, and Hilbert-space scaling derives the counts. Particle slots remain useful coordinate arguments; they are not persistent identities of identical particles.

Fixed N versus Fock Space

A fixed-particle calculation needs only one sector $\mathcal H_N^{(\pm)}$. Fock space collects all allowed particle-number sectors:

$$\mathcal F_\pm(\mathcal h) = \bigoplus_{N=0}^{\infty} \mathcal H_N^{(\pm)}.$$

A Fock-space vector is a sector sequence

$$\lvert\Psi\rangle = \bigoplus_{N=0}^{\infty} \lvert\psi_N\rangle,$$

with

$$\sum_{N=0}^{\infty} \lVert\psi_N\rVert^2 < \infty.$$

This direct sum has a different structural meaning from a tensor product. Values of $N$ label orthogonal alternatives, not simultaneously present subsystems. Direct sums versus tensor products explains this distinction.

When Fock space is useful

Fock space is natural when:

• creation and annihilation operators are used;
• a reservoir exchanges particles with the system;
• pairing or source terms connect number sectors;
• a grand-canonical ensemble is convenient;
• particles are excitations of modes rather than permanently labeled objects;
• one wants one operator language that acts consistently across all fixed-$N$ sectors.

Fock space does not imply that the physical state has indefinite particle number. A number eigenstate belongs to Fock space but has support in only one sector. A number-conserving Hamiltonian is block diagonal on Fock space and can be restricted to a fixed-$N$ block.

See Fock space for the Hilbert direct sum and the domains of number-changing operators.

Local Tensor-Product Systems

Not every many-body problem begins with identical particles. A spin model assigns a local Hilbert space $\mathcal h_i$ to each distinguishable site $i$. For $L$ sites,

$$\mathcal H_{\mathrm{lat}} = \bigotimes_{i=1}^{L} \mathcal h_i.$$

For a spin-$s$ degree of freedom,

$$\dim\mathcal h_i = 2s+1.$$

A spin-$1/2$ chain therefore has

$$\mathcal H_L = (\mathbb C^2)^{\otimes L}, \qquad \dim\mathcal H_L = 2^L.$$

The tensor factors are sites, not particle labels. A local spin flip acts on one site factor. Exchange statistics is not imposed by symmetrizing the site tensor product.

This distinction matters for entanglement. A spatial cut of a spin chain naturally partitions site factors. An identical-particle problem requires a declared mode, region, or observable-algebra partition before an entanglement claim has operational meaning.

Lattice Particles and Equivalent Finite Constructions

Particle and site viewpoints can describe the same finite lattice model in different languages.

For a spinful fermion orbital at one site, the local states are

$$\lvert0\rangle, \quad \lvert\uparrow\rangle, \quad \lvert\downarrow\rangle, \quad \lvert\uparrow\downarrow\rangle.$$

The local space has dimension four, so $L$ sites give

$$\mathcal H_{\mathrm{site}} = (\mathbb C^4)^{\otimes L}, \qquad \dim\mathcal H_{\mathrm{site}} = 4^L.$$

The same finite system can be described as the fermionic Fock space of $2L$ ordered spin-orbitals:

$$\mathcal F_-(\mathbb C^{2L}), \qquad \dim\mathcal F_-(\mathbb C^{2L}) = 2^{2L} = 4^L.$$

The vector-space dimensions match, but fermionic operator ordering and graded signs must be handled consistently. A naive site tensor product does not erase the anticommutation algebra. Jordan–Wigner strings in one dimension are one way those signs reappear when fermions are represented by ordinary spin operators.

For lattice bosons, each site has infinitely many occupation states unless a local cutoff $n_i\le n_{\max}$ is imposed. With that numerical cutoff,

$$\mathcal H_{\mathrm{trunc}} = \bigotimes_{i=1}^{L} \mathbb C^{n_{\max}+1},$$

and

$$\dim\mathcal H_{\mathrm{trunc}} = (n_{\max}+1)^L.$$

This is generally a computational approximation. But in a fixed-total-number sector, $n_{\max}\ge N$ excludes no allowed occupation, so that local cutoff is exact within the sector. A microscopic hard-core constraint is another, physically distinct finite local space.

Symmetry Sectors and Conserved Blocks

Suppose a self-adjoint operator $Q$ has spectral subspaces $\mathcal H_q$. In a discrete finite-dimensional setting,

$$\mathcal H = \bigoplus_q \mathcal H_q.$$

If

$$[H,Q] = 0,$$

then the Hamiltonian preserves each sector:

$$H\mathcal H_q \subseteq \mathcal H_q.$$

Writing $P_q$ for the projector onto $\mathcal H_q$,

$$H = \bigoplus_q H_q, \qquad H_q = P_qHP_q.$$

Typical labels include particle number, magnetization, crystal momentum, parity, total angular momentum, and irreducible representation labels.

Working in one sector has three benefits:

• it enforces the desired physical constraint exactly;
• it reduces the matrix dimension;
• it prevents numerical mixing between quantum numbers that the exact dynamics conserves.

Not every useful basis simultaneously diagonalizes every symmetry. Noncommuting conserved quantities require compatible labels, such as total $S^2$ together with one spin component $S_z$.

Constraints Are More Than Convenient Blocks

Some physical state spaces are kernels or projected subspaces defined by constraints:

$$\mathcal H_{\mathrm{phys}} = \left\{ \lvert\Psi\rangle\in\mathcal H_{\mathrm{kin}}: G_a\lvert\Psi\rangle=0 \text{ for every }a \right\}.$$

Examples include hard constraints on local occupancy, fixed total charge, gauge constraints, and effective models that exclude high-energy configurations.

Three ideas should be separated:

1. a conserved sector is preserved by a specified Hamiltonian;
2. a physical constraint declares some kinematic vectors unphysical in the model;
3. a superselection restriction limits accessible coherence or observables between sectors.

The equation $[H,Q]=0$ proves conservation under that Hamiltonian. It does not by itself prove a universal superselection rule. Particle-Number Superselection Preview explains the operational distinction.

Representations Do Not Change the Physical Space

A Hilbert space can be described in coordinate, momentum, occupation, or energy-eigenstate representations. These choices change the coordinates of states and operators, not the underlying physical state space when the transformation is unitary and complete.

For a fixed-$N$ state,

$$\Psi(x_1,\ldots,x_N) = \langle x_1,\ldots,x_N\vert\Psi\rangle$$

is its coordinate representation. In a mode basis,

$$\lvert\Psi\rangle = \sum_{\boldsymbol n} C_{\boldsymbol n} \lvert\boldsymbol n\rangle.$$

In a field-operator language, the same state can be constructed by applying smeared creation fields to the vacuum. A tensor network is instead a parameterization of state amplitudes. A fixed bond-dimension family generally forms a nonlinear subset, not a basis or a linear Hilbert subspace: adding two such states can require a larger bond dimension. Schollwöck’s review develops this distinction for matrix product states.

The representations emphasize different structures:

Language	Natural labels	Best suited to
first-quantized wavefunction	particle slots and one-particle coordinates	fixed small $N$, coordinate-space boundary conditions
occupation-number basis	mode populations	exchange symmetry, sparse operators, exact diagonalization
momentum-space basis	mode momenta and internal labels	translation invariance, Fermi surfaces, scattering
real-space fields	position and internal labels	local densities, continuum interactions, correlations
local tensor basis	site configurations	spin systems, lattice locality, tensor networks

First-Quantized Many-Body Wavefunctions constructs coordinate amplitudes. Occupation-Number Representation constructs mode bases for sparse calculations. Field Operators in Many-Body Models develops the continuum operator language.

Worked Example: Two Spin-1/2 Fermions in Two Orbitals

Let the spatial orbitals be $a$ and $b$. Including spin, the one-particle modes are

$$a\uparrow, \quad a\downarrow, \quad b\uparrow, \quad b\downarrow.$$

Thus

$$\dim\mathcal h = 4.$$

The two-fermion sector has dimension

$$\dim\bigwedge^2\mathcal h = \binom42 = 6.$$

Two basis states have double occupancy of one spatial orbital,

$$\lvert a\uparrow,a\downarrow\rangle, \qquad \lvert b\uparrow,b\downarrow\rangle,$$

and four have one fermion in each orbital. Recombining the latter into total-spin singlet and triplet combinations is another basis change inside the same six-dimensional space.

This example shows why “two orbitals” does not mean $M=2$ when spin is an active one-particle label.

Operators Must Match the Space

An operator is not specified by a symbolic formula alone. Its domain, codomain, statistics, and sector action matter.

For a one-particle operator $A$ with matrix elements $A_{ij}$, its number-conserving lift to Fock space has the pattern

$$d\Gamma(A) = \sum_{ij} A_{ij} a_i^\dagger a_j.$$

It preserves particle number because each term annihilates and creates one particle. A pairing term such as

$$\Delta_{ij} a_i^\dagger a_j^\dagger + \Delta_{ij}^* a_j a_i$$

connects sectors whose particle numbers differ by two and therefore cannot act within one fixed-$N$ space by itself.

On a site tensor product, a local operator is embedded as

$$O_i = I_1\otimes\cdots\otimes I_{i-1} \otimes O \otimes I_{i+1}\otimes\cdots\otimes I_L.$$

The One-Body Operators and Two-Body Operators application guides develop matrix elements, reduced density matrices, interaction conventions, and computational checks. This overview uses those structures only to show how operator form reveals the required Hilbert space.

Pure States, Mixed States, and Ensembles

Choosing a Hilbert space does not choose a state. A pure state is a ray represented by $\lvert\Psi\rangle\in\mathcal H$. A density operator $\rho$ on that space is positive and has trace one. It represents a pure state when its rank is one, and a mixed state otherwise.

When the Gibbs trace is finite, fixed-number equilibrium has

$$\rho_N = \frac{e^{-\beta H_N}}{Z_N}, \qquad Z_N = \operatorname{Tr}_{\mathcal H_N} e^{-\beta H_N}.$$

For a grand-canonical description on Fock space,

$$\rho_{\mathrm{GC}} = \frac{ e^{-\beta(H-\mu N)} }{ \Xi },$$

with

$$\Xi = \operatorname{Tr}_{\mathcal F} e^{-\beta(H-\mu N)}.$$

If $[H,N]=0$, this density operator is block diagonal in particle number. Grand-canonical number fluctuations do not require coherent pure-state superpositions of different $N$; they can arise from a classical mixture of number sectors.

The Canonical Ensemble and Grand Canonical Ensemble pages derive these states and their equivalence conditions.

Truncation Is Part of the Model Statement

Continuum and unrestricted bosonic Hilbert spaces are infinite dimensional. Numerical work often retains a finite subspace, or a smaller variational family of states. Common cutoffs include:

• a finite spatial volume, which discretizes momenta but does not alone cap their number;
• a finite one-particle basis of $M$ modes;
• a momentum or energy cutoff;
• a maximum local boson occupation $n_{\max}$;
• a maximum total excitation or particle number;
• a selected symmetry sector, exact for dynamics that stays in it;
• an active-space restriction around a reference configuration.

Let $P_\Lambda$ project onto the retained subspace. The simplest projected Hamiltonian is

$$H_\Lambda = P_\Lambda H P_\Lambda.$$

This expression does not guarantee controlled accuracy. Eliminated states can renormalize couplings, generate new operators, or produce energy-dependent effective interactions. A trustworthy calculation states:

1. what $\Lambda$ means;
2. which observables are expected to be insensitive to it;
3. how convergence is tested as the retained space grows;
4. whether couplings have been matched or renormalized at that cutoff;
5. which symmetries the truncation preserves or breaks.

A local occupation cutoff that changes a result when increased is not a harmless implementation detail. It is an unresolved approximation error.

Basis Changes and Truncations Do Not Commute Automatically

Suppose $U$ is a unitary basis change on the full one-particle space. If $P_M$ selects $M$ retained modes, then in general

$$P_MU \ne UP_M.$$

Truncating localized orbitals and truncating energy eigenmodes can therefore produce different finite models even when the complete bases span the same infinite-dimensional space.

If the same physical retained subspace is carried through the unitary transformation, its projector becomes $P'_M=UP_MU^\dagger$, and the projected Hamiltonians are exactly related:

$$P'_M(UHU^\dagger)P'_M=U(P_MHP_M)U^\dagger.$$

The mismatch arises when one chooses different physical cutoffs in the two bases, not from a failure of a consistent change of coordinates. Distinguish:

• a unitary representation change in a complete space;
• a projection to a finite active subspace;
• an effective model obtained after integrating out excluded states.

Only the first is automatically an exact reformulation.

Finite Volume and the Thermodynamic Limit

At finite volume or finite lattice size, one can usually specify an ordinary Hilbert space $\mathcal H_L$ and Hamiltonian $H_L$. The thermodynamic limit is a sequence

$$(\mathcal H_L,H_L) \quad\text{with}\quad L,N,V\to\infty$$

under a declared scaling, such as fixed density.

It is often misleading to imagine that every infinite system is simply one enormous tensor product with all finite-volume vectors embedded uniquely. In the infinite limit:

• physically relevant states may be defined through expectation values of local observables;
• distinct phases can lead to inequivalent Hilbert-space representations;
• symmetry-broken states can emerge only after the order of limits and source removal is specified;
• thermal states need not be represented by trace-class density matrices on a naive infinite-volume space.

Finite-volume calculations remain the controlled starting point. Thermodynamic limits explains boundary effects and the order of limits. Operator-algebraic infinite-system constructions are an advanced extension, not assumed silently in finite-model formulas.

Locality Depends on the Factorization

An operator is local only relative to a declared decomposition.

For a lattice tensor product, “local” commonly means support on a bounded set of sites. In a continuum field theory, it means an operator density smeared over a spatial region. In a momentum basis, the same real-space local interaction may connect many momenta. In an energy eigenbasis, even a simple local observable can be dense.

Similarly, entanglement depends on a subsystem or algebra decomposition. Exchange-symmetrized particle slots are not automatically independent parties. A mode partition, spatial partition, species partition, or site partition must be stated.

This dependence is not a weakness. It records which operations and measurements are physically local in the problem being modeled.

Exercises

Identify the construction

Choose the natural starting space for each system.

1. Three distinguishable trapped ions, retaining only two internal levels per ion.
2. Four identical spinless bosons in six orbitals.
3. Four identical spinless fermions in six orbitals.
4. A chain of ten spin-$1/2$ moments.
5. A spinful electron lattice in which particle number may fluctuate.

Solution

1. In the retained internal-state model, the ions are distinguishable local objects, so

$$(\mathbb C^2)^{\otimes3}.$$

2. Use the fixed bosonic sector

$$\operatorname{Sym}^4\mathbb C^6.$$

3. Use the fixed fermionic sector

$$\bigwedge^4\mathbb C^6.$$

4. The sites are distinguishable factors:

$$(\mathbb C^2)^{\otimes10}.$$

5. Use fermionic Fock space over the retained spin-orbitals. An ordered four-state local basis per site is an equivalent finite representation when its fermionic signs are implemented consistently.

A cutoff convergence test

A bosonic lattice calculation uses local occupations $0\le n_i\le n_{\max}$. State a minimal convergence protocol for a low-energy observable $O$.

Solution

At fixed physical parameters and system size:

1. compute $O(n_{\max})$ for a sequence of increasing cutoffs;
2. monitor the probability weight in the highest retained occupation states;
3. compare energy levels and other observables sensitive to local density;
4. verify that the desired tolerance is stable under at least one further cutoff increase;
5. repeat the check in the most strongly occupied parameter regime used in the study.

A small change at one parameter point does not prove uniform convergence across a phase diagram. If the interaction has been matched at a cutoff, that matching prescription must be held consistent as $n_{\max}$ changes.

Why the thermodynamic limit is not one finite trace

Explain why the formula

$$\rho_L = \frac{e^{-\beta H_L}}{Z_L}$$

at every finite $L$ does not automatically define a trace-class density operator by simply setting $L=\infty$.

Solution

Both the Hilbert space and Hamiltonian change with $L$. The partition function typically grows exponentially with volume, and an infinite-volume Gibbs state is characterized by consistent expectation values of local observables rather than by a naive trace over one infinite tensor product.

Different limiting states can also arise from boundary conditions, infinitesimal sources, or distinct phases. A controlled thermodynamic limit therefore specifies a sequence of finite systems and tests convergence of local or intensive quantities. Writing $L=\infty$ inside a finite-volume trace suppresses these structural choices.

What projection omits

Let $H=\begin{pmatrix}0&t\\t&\Delta\end{pmatrix}$ with $\Delta>0$ and real $|t|\ll\Delta$. Retain only the first basis vector. Compare $PHP$ with the exact lower eigenvalue and explain the difference.

Solution

The one-dimensional projected Hamiltonian is zero. The exact lower energy is

$$E_-=\frac{\Delta-\sqrt{\Delta^2+4t^2}}2 =-\frac{t^2}{\Delta}+O(t^4/\Delta^3).$$

Virtual admixture of the omitted state lowers the energy. A matched low-energy effective Hamiltonian includes this correction, whereas simple projection does not. The retained state can have nearly unit overlap with the exact ground state while its projected energy still misses the leading nonzero shift.

References

• A. Altland and B. Simons, Condensed Matter Field Theory, 2nd ed. (Cambridge University Press, 2010) — many-body state spaces, field operators, and statistical ensembles.
• F. A. Berezin, The Method of Second Quantization (Academic Press, 1966) — mathematical operator and Fock-space formalism.
• O. Bratteli and D. W. Robinson, Operator Algebras and Quantum Statistical Mechanics 2, 2nd ed. (Springer, 1997) — equilibrium states and infinite quantum systems.
• P. Coleman, Introduction to Many-Body Physics (Cambridge University Press, 2015) — model spaces, second quantization, lattices, and emergent many-body descriptions.
• A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems (McGraw–Hill, 1971; Dover, 2003) — fixed-number and second-quantized formulations of nonrelativistic many-body theory.
• J. W. Negele and H. Orland, Quantum Many-Particle Systems (Addison–Wesley, 1988) — Fock space, coherent-state methods, and finite-temperature many-body theory.
• M. Reed and B. Simon, Methods of Modern Mathematical Physics, Vol. I: Functional Analysis, revised ed. (Academic Press, 1980) — Hilbert spaces, tensor products, operators, and spectral subspaces.
• U. Schollwöck, “The density-matrix renormalization group in the age of matrix product states”, Annals of Physics 326, 96–192 (2011) — local tensor-product spaces, controlled state approximations, and one-dimensional many-body computation.