Scaling of Hilbert Space

The first quantitative sign of a many-body problem is often not an interaction strength or a length scale. It is the number of independent basis states.

For a one-particle Hilbert space of dimension $M$, adding particles or sites can produce state spaces whose dimensions grow exponentially or combinatorially. The exact count depends on what is physically distinguishable, which particle statistics apply, what conserved quantities are fixed, and which truncations have been imposed.

This page is the canonical home for those counts and their asymptotic interpretation. The Many-Body Hilbert Spaces Overview owns the modeling choice among fixed-number sectors, Fock space, local tensor products, and constrained blocks. The construction of tensor products, exchange-symmetric subspaces, and Fock space lives in the linked prerequisite pages. Here the goal is to answer:

1. What is the many-body Hilbert-space dimension?
2. Which physical assumptions determine that dimension?
3. How rapidly does it grow?
4. How much can symmetries and constraints reduce it?
5. What does the count imply, and what does it not imply, about computation and physics?

Count the Physical Space, Not the Symbols

Before using a formula, identify four ingredients.

Question	Typical choices
What are the elementary factors or modes?	sites, spins, orbitals, momentum modes
Are particles distinguishable?	labeled subsystems, bosons, fermions
What is fixed?	particle number, magnetization, charge, momentum
What is truncated?	number of modes, local occupation, total excitation

The same notation can conceal different spaces. For example, $M$ orbitals occupied by $N$ particles may mean:

• $N$ labeled particles, giving an $M^N$ tensor-product space;
• $N$ bosons, giving a symmetric fixed-number sector;
• $N$ spinless fermions, giving an antisymmetric fixed-number sector;
• all possible fermion numbers, giving a full fermionic Fock space;
• bosons with a numerical occupation cutoff, giving a truncated space.

These are not alternative ways to count one object. They are different physical or computational state spaces.

Distinguishable Local Degrees of Freedom

Suppose a lattice has $L$ distinguishable sites and site $i$ has a local Hilbert space $\mathcal h_i$ of dimension $d_i$. The full space is

$$\mathcal H_L = \bigotimes_{i=1}^{L} \mathcal h_i,$$

so

$$\dim\mathcal H_L = \prod_{i=1}^{L} d_i.$$

For identical local dimensions $d_i=d$,

$$D_L \equiv \dim\mathcal H_L = d^L.$$

Each added site multiplies the state-space dimension:

$$\frac{D_{L+1}}{D_L} = d.$$

This multiplication is the source of exponential growth. It follows from the tensor-product structure before a Hamiltonian has been specified.

Spin systems

A spin-$s$ degree of freedom has local dimension

$$d=2s+1.$$

Therefore $L$ spin-$s$ sites span

$$D_L = (2s+1)^L.$$

For spin-$1/2$,

$$D_L=2^L.$$

A generic pure state then requires $2^L$ complex amplitudes in a product basis:

$$|\psi\rangle = \sum_{\sigma_1,\ldots,\sigma_L} c_{\sigma_1\cdots\sigma_L} |\sigma_1,\ldots,\sigma_L\rangle.$$

Normalization and an overall phase reduce the number of independent real parameters slightly, but not the exponential scaling.

Heterogeneous local spaces

Not every site needs the same dimension. A register containing $L$ qubits and one qutrit has

$$D = 3\,2^L.$$

A spin chain coupled to a locally truncated oscillator with states $n=0,\ldots,n_{\max}$ has

$$D = 2^L(n_{\max}+1).$$

Writing the product $\prod_i d_i$ first prevents an inappropriate use of $d^L$ when local cutoffs differ.

Operators and Density Matrices Grow Faster

If $\dim\mathcal H=D$, the vector space of linear operators on $\mathcal H$ has dimension

$$\dim\mathcal B(\mathcal H) = D^2.$$

A dense operator or density matrix therefore has $D^2$ complex entries before Hermiticity, positivity, or trace constraints are used. For $L$ qubits,

$$D=2^L, \qquad D^2=4^L.$$

This count is why explicitly storing a generic mixed state becomes prohibitive at smaller $L$ than storing a state vector.

The statement concerns a generic dense representation. A local Hamiltonian can often be stored as a short sum of local terms even though its matrix acts on a space of dimension $D$.

Labeled Particles in a Finite One-Particle Space

Let the one-particle space have dimension $M$. For $N$ distinguishable particles,

$$\mathcal H_N = \mathcal h^{\otimes N}, \qquad \dim\mathcal H_N = M^N.$$

A product basis is labeled by an ordered list

$$(i_1,i_2,\ldots,i_N), \qquad i_a\in\{1,\ldots,M\}.$$

The ordering matters because particle labels matter. The basis states $(1,2)$ and $(2,1)$ are distinct for two labeled particles even if the same two one-particle modes appear.

For identical particles, those ordered lists overcount physical states. Exchange symmetry changes the counting.

Fixed-Number Bosonic Sectors

Place $N$ identical bosons in $M$ one-particle modes. A number state is specified by nonnegative occupations

$$(n_1,\ldots,n_M), \qquad \sum_{i=1}^{M} n_i=N.$$

The number of solutions is the stars-and-bars count

$$\begin{aligned} D_{\mathrm B}(M,N) &= \binom{M+N-1}{N} \\ &= \binom{M+N-1}{M-1}. \end{aligned}$$

One way to see the formula is to represent the $N$ particles by stars and divide the sequence into $M$ mode occupancies using $M-1$ separators. There are $M+N-1$ positions and one chooses the locations of either the $N$ stars or the $M-1$ separators.

Example: four bosons in eight modes

For $M=8$ and $N=4$,

$$\begin{aligned} D_{\mathrm B}(8,4) &= \binom{11}{4} \\ &= 330. \end{aligned}$$

The corresponding labeled-particle space would have dimension

$$8^4=4096.$$

Symmetrization substantially reduces the count, but the bosonic sector remains large.

Fixed-Number Fermionic Sectors

For $N$ identical spinless fermions in $M$ modes, each occupation obeys

$$n_i\in\{0,1\}.$$

A basis state is therefore a choice of $N$ occupied modes from $M$:

$$D_{\mathrm F}(M,N) = \binom{M}{N}.$$

If $N>M$, the dimension is zero. This is the finite-mode expression of Pauli exclusion.

For four spinless fermions in eight modes,

$$D_{\mathrm F}(8,4) = \binom{8}{4} = 70.$$

The fermionic sector is smaller than the bosonic sector for the same $M$ and $N$ because repeated mode occupation is forbidden.

Several Species and Spinful Fermions

If up-spin and down-spin particle numbers are separately fixed in $L$ spatial orbitals, the sectors factor:

$$D(L,N_\uparrow,N_\downarrow) = \binom{L}{N_\uparrow} \binom{L}{N_\downarrow}.$$

The product appears because one chooses the occupied spatial orbitals independently for each spin species. Double occupation of a spatial orbital is allowed when the two particles have opposite spin.

For a Hubbard chain at half filling in the sector

$$N_\uparrow = N_\downarrow = \frac{L}{2},$$

the dimension is

$$D_{\mathrm{Hub}} = \binom{L}{L/2}^{\!2}.$$

If only the total particle number $N=N_\uparrow+N_\downarrow$ is fixed, there are $2L$ spin-orbitals and

$$D(L,N) = \binom{2L}{N}.$$

Fixing more commuting charges selects a smaller block. It does not change the dimension of the full physical Hilbert space.

Full and Truncated Fock Spaces

Fock space is a direct sum over particle-number sectors. Its dimension depends strongly on particle statistics and on truncation.

Fermions

Each of $M$ fermionic modes is either empty or occupied. The full Fock-space dimension is

$$\dim\mathcal F_{\mathrm F} = \sum_{N=0}^{M} \binom{M}{N} = 2^M.$$

This is the binomial theorem. A system of $M$ fermionic modes has the same dimension as $M$ qubits, although operator signs and physical interpretations differ.

Bosons

Each bosonic mode can have arbitrarily large occupation, so the full bosonic Fock space over any nonzero finite set of modes is infinite dimensional.

A finite numerical basis therefore requires an explicit cutoff.

If the total particle number is restricted to $N\leq N_{\max}$, then

$$\begin{aligned} D_{\mathrm B}^{\mathrm{total}} (M,N_{\max}) &= \sum_{N=0}^{N_{\max}} \binom{M+N-1}{N} \\ &= \binom{M+N_{\max}}{N_{\max}}. \end{aligned}$$

If instead each mode has the local cutoff

$$0\leq n_i\leq n_{\max},$$

then

$$D_{\mathrm B}^{\mathrm{local}} = (n_{\max}+1)^M.$$

These truncations are not equivalent. The local cutoff allows total occupation as large as $Mn_{\max}$, while the total-number cutoff excludes every state above $N_{\max}$.

Continuous One-Particle Spaces

A particle in continuous space has an infinite-dimensional one-particle Hilbert space. A finite combinatorial count arises only after choosing a finite basis, spatial grid, energy cutoff, momentum cutoff, or other regulator.

For example, placing $N$ fermions into a numerical basis of $M$ orbitals gives

$$D_{\mathrm F}(M,N) = \binom{M}{N},$$

but this is the dimension of the regulated model, not of the exact continuum Hilbert space.

A trustworthy numerical statement therefore reports:

• the one-particle basis;
• the number $M$ of retained modes;
• the particle-number and symmetry sector;
• every local or total occupation cutoff;
• convergence as the cutoff is enlarged.

Without those data, a reported Hilbert-space dimension is not reproducible.

Conserved Charges and Block Decomposition

If

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

then $\mathcal H$ decomposes into invariant eigenspaces of $Q$:

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

The Hamiltonian is block diagonal:

$$H = \bigoplus_q H_q.$$

Working in one sector can reduce storage and computation dramatically. It is essential, however, to distinguish:

• the dimension of the full physical space;
• the dimension of a conserved sector;
• the dimension of an irreducible symmetry block;
• the dimension of a variational or truncated ansatz.

These numbers answer different questions.

Magnetization sectors

For $L$ spin-$1/2$ sites, fixing the number $N_\uparrow$ of up spins gives

$$D(L,N_\uparrow) = \binom{L}{N_\uparrow}.$$

Equivalently, this fixes the total magnetization

$$S^z_{\mathrm{tot}} = \hbar \left( N_\uparrow-\frac{L}{2} \right).$$

At half filling, $N_\uparrow=L/2$ for even $L$.

Translation and point-group sectors

Translation, inversion, reflection, and point-group symmetries can split a fixed-charge sector further. A rough estimate might divide a generic sector dimension by the number of symmetry labels, but exact dimensions depend on fixed configurations and orbit lengths. One should construct the symmetry representation or count its characters rather than assume equal blocks.

Fully permutation-symmetric spins

The fully symmetric subspace of $L$ spin-$1/2$ particles has total spin $S=L/2$ and dimension

$$D_{\mathrm{sym}} = 2S+1 = L+1.$$

This polynomial dimension is far smaller than $2^L$. It applies only when the state and dynamics remain in that collective-spin sector. It is not the dimension of a generic spin chain.

Asymptotic Growth and Entropy Density

Exact integers are useful for finite calculations. Asymptotic formulas reveal the growth class.

For a family of spaces with dimension $D_L$, define a Hilbert-space growth density when the limit exists:

$$s_{\mathcal H} = \lim_{L\to\infty} \frac{1}{L} \ln D_L.$$

This quantity resembles an entropy density because $\ln D_L$ counts the number of basis states on a logarithmic scale. It is not automatically a thermodynamic entropy of a physical state.

For $d$-level sites,

$$s_{\mathcal H} = \ln d.$$

Binomial sectors at fixed filling

Let

$$N=pL, \qquad 0<p<1.$$

Stirling’s approximation gives

$$\binom{L}{pL} \sim \frac{ e^{L h(p)} }{ \sqrt{2\pi Lp(1-p)} },$$

where

$$h(p) = -p\ln p - (1-p)\ln(1-p).$$

The function $h(p)$ is the binary entropy in natural logarithmic units. At $p=1/2$,

$$\binom{L}{L/2} \sim \frac{2^L}{\sqrt{\pi L/2}}.$$

The conserved sector is smaller than the full $2^L$ space by a factor proportional to $\sqrt L$, but it is still exponentially large.

Taking the logarithm makes the leading and subleading behavior explicit:

$$\ln \binom{L}{L/2} = L\ln2 - \frac{1}{2}\ln L + O(1).$$

Fermions at fixed filling

For $N=\nu M$ spinless fermions,

$$\ln D_{\mathrm F}(M,\nu M) = M h(\nu) + O(\ln M).$$

The sector grows exponentially in the number of modes whenever $0<\nu<1$ is held fixed.

Bosons at fixed density per mode

For $N=\nu M$ bosons,

$$\begin{aligned} \ln D_{\mathrm B}(M,\nu M) ={}& M(1+\nu)\ln(1+\nu) \\ &- M\nu\ln\nu + O(\ln M). \end{aligned}$$

Thus exchange symmetrization does not generally turn a fixed-density bosonic problem into a polynomial-size problem.

Concrete Representation Costs

Suppose one complex amplitude is stored in 16 bytes using two 64-bit floating-point numbers. A dense spin-$1/2$ state vector then requires approximately

$$\text{memory} = 16\times2^L \ \text{bytes}.$$

Sites $L$	Amplitudes $2^L$	Dense state-vector memory
20	$1.05\times10^6$	$16$ MiB
30	$1.07\times10^9$	$16$ GiB
40	$1.10\times10^{12}$	$16$ TiB
50	$1.13\times10^{15}$	$16$ PiB

These estimates omit temporary vectors, operator storage, indexing, and algorithmic overhead. An eigensolver often needs several state-sized work arrays.

A dense $L=20$ qubit density matrix already contains

$$4^{20} = 2^{40}$$

complex entries and would require about $16$ TiB in the same representation.

The units matter. Here MiB, GiB, TiB, and PiB denote powers of $2^{20}$, $2^{30}$, $2^{40}$, and $2^{50}$ bytes.

Locality Does Not Shrink the Hilbert Space

A local Hamiltonian may contain only order-$L$ terms:

$$H = \sum_X h_X,$$

with each $h_X$ supported on a small region. This structure can make matrix-vector products sparse and enables specialized algorithms. It does not change

$$\dim\mathcal H_L=d^L.$$

Locality in Many-Body Systems constrains operator support, interaction range, and propagation; it does not change the kinematic state-space count.

Similarly, a Hamiltonian matrix can be sparse while its eigenstates are highly entangled and require exponentially many coefficients in the chosen product basis.

Exponential Dimension Is Not an Algorithmic Theorem

Hilbert-space growth is an unconditional counting fact. It does not by itself prove that every observable, ground state, or dynamical question requires exponential time.

Important structured exceptions include:

• product states, specified by order-$L$ local data;
• Gaussian bosonic or fermionic states, specified by polynomial-size covariance data;
• permutation-symmetric collective states;
• stabilizer states;
• low-entanglement states represented by matrix product states, previewed alongside other structured ansätze in Variational Many-Body States;
• exactly solvable models;
• observables accessible by Monte Carlo sampling when a useful nonnegative representation exists.

Conversely, a compact Hamiltonian description does not guarantee an easy solution. One must distinguish:

1. the size of the ambient Hilbert space;
2. the size of the physically relevant sector;
3. the complexity of representing the target state;
4. the complexity of computing the chosen observable;
5. worst-case complexity of a problem class.

These quantities can scale differently.

Basis Size Is Not Physical Entropy

The logarithm of a Hilbert-space dimension is a capacity:

$$\ln D$$

counts how many orthogonal basis states are available. A particular pure state can have zero von Neumann entropy even when it lives in an exponentially large space:

$$S(|\psi\rangle\langle\psi|) = 0.$$

An equal mixture over a $D$-dimensional space has

$$\rho = \frac{I_D}{D}, \qquad S(\rho) = k_{\mathrm B}\ln D.$$

The equality in the second case depends on the state being maximally mixed. One should not call $\ln D$ the entropy of an arbitrary state.

Entanglement entropy is different again: it depends on a subsystem decomposition and the reduced state, not merely on the dimension of the full space.

Worked Comparison

Consider six particles and twelve one-particle modes.

Distinguishable particles

For labeled particles,

$$D_{\mathrm{dist}} = 12^6 = 2{,}985{,}984.$$

Bosons

For identical bosons,

$$\begin{aligned} D_{\mathrm B} &= \binom{17}{6} \\ &= 12{,}376. \end{aligned}$$

Spinless fermions

For identical spinless fermions,

$$\begin{aligned} D_{\mathrm F} &= \binom{12}{6} \\ &= 924. \end{aligned}$$

The three dimensions differ by orders of magnitude because they represent three different exchange assumptions. None can be selected from $M$ and $N$ alone.

A Reliable Counting Workflow

1. Define the one-body or local basis. State every mode, site, spin label, and cutoff.
2. Specify distinguishability and statistics. Decide whether tensor-product labels, symmetric states, or antisymmetric states are physical.
3. Fix superselection and conserved sectors. State particle number, magnetization, charge, momentum, and parity conditions.
4. Apply basis truncations. Distinguish local occupation cutoffs from total-number or energy cutoffs.
5. Count before constructing. Estimate exact dimension and asymptotic growth before assembling a matrix.
6. Report the represented space. A computational paper should identify the full space and the smaller block actually used.
7. Test cutoff convergence. A finite basis approximating an infinite-dimensional problem must be enlarged until target observables stabilize.

Canonical Boundaries

This page owns comparative dimension formulas, their asymptotics, and their implications for many-body representation cost.

The linked pages retain the following canonical roles:

• tensor-product pages construct composite Hilbert spaces and product bases;
• identical-particle pages derive exchange symmetry and symmetrization;
• Fock-space pages construct particle-number sectors and occupation-number states;
• numerical-mathematics pages explain sparse storage and eigensolvers;
• Computational QM owns reusable algorithm implementations and benchmarks;
• later entanglement pages explain when structured state representations avoid generic dense storage.

Common Mistakes

• Confusing the dimension $M$ of the one-particle space with the many-body dimension.
• Using $M^N$ for identical particles without imposing exchange symmetry.
• Using $\binom{M}{N}$ for bosons or $\binom{M+N-1}{N}$ for fermions.
• Forgetting that the full bosonic Fock space is infinite dimensional without a cutoff.
• Failing to say whether a bosonic cutoff is local, total-number, energy, or mode based.
• Reporting a symmetry-sector dimension as though it were the full Hilbert-space dimension.
• Assuming every symmetry block has exactly the same size.
• Treating a sparse Hamiltonian as evidence that generic state vectors are sparse.
• Equating $\ln D$ with the entropy of every state in a $D$-dimensional space.
• Concluding from exponential dimension alone that every physical prediction has exponential computational cost.
• Quoting a continuum basis size without a cutoff-convergence test.
• Forgetting internal species when counting spinful fermions.

Exercises

Mixed local dimensions

A system contains ten qubits, two spin-$1$ sites, and one oscillator truncated to occupations $n=0,\ldots,4$. Find the full tensor-product dimension.

Solution

Each qubit contributes dimension $2$, each spin-$1$ site contributes dimension $3$, and the truncated oscillator contributes dimension $5$. Therefore

$$\begin{aligned} D &= 2^{10}\,3^2\,5 \\ &= 46{,}080. \end{aligned}$$

The count assumes all listed degrees of freedom are simultaneously present and that no conserved sector has been selected.

Bosons, fermions, and labels

Five particles occupy nine one-particle modes. Find the dimensions for labeled particles, identical bosons, and identical spinless fermions.

Solution

For labeled particles,

$$D_{\mathrm{dist}} = 9^5 = 59{,}049.$$

For bosons,

$$\begin{aligned} D_{\mathrm B} &= \binom{9+5-1}{5} \\ &= \binom{13}{5} \\ &= 1287. \end{aligned}$$

For spinless fermions,

$$D_{\mathrm F} = \binom{9}{5} = 126.$$

The differences arise from distinguishability and allowed mode occupations.

Half-filled spin sector

Use Stirling’s approximation to show that fixing $N_\uparrow=L/2$ does not remove exponential growth from an even spin-$1/2$ chain.

Solution

The sector dimension is

$$D_{L/2} = \binom{L}{L/2} = \frac{L!}{[(L/2)!]^2}.$$

Using

$$n! \sim \sqrt{2\pi n} \left( \frac{n}{e} \right)^n$$

gives

$$D_{L/2} \sim \frac{2^L}{\sqrt{\pi L/2}}.$$

The reduction relative to $2^L$ is only a factor proportional to $\sqrt L$. The logarithm still has the extensive leading term:

$$\ln D_{L/2} = L\ln2 - \frac{1}{2}\ln L + O(1).$$

State-vector memory

Assume 16 bytes per complex amplitude. Estimate the dense memory for a 36-site spin-$1/2$ state vector in GiB. How much is required in the zero-magnetization sector?

Solution

The full vector requires

$$\begin{aligned} M_{\mathrm{full}} &= 16\times2^{36}\ \text{bytes} \\ &= 2^{40}\ \text{bytes} \\ &= 1024\ \text{GiB}. \end{aligned}$$

The zero-magnetization sector has

$$\binom{36}{18} = 9{,}075{,}135{,}300$$

amplitudes, requiring approximately

$$\frac{ 16\binom{36}{18} }{ 2^{30} } \approx 135.2\ \text{GiB}.$$

The conserved sector helps substantially, but the resulting vector is still large and an iterative solver would need additional work arrays.

Compare bosonic cutoffs

For $M=4$ bosonic modes, compare a total-number cutoff $N\leq3$ with a local cutoff $0\leq n_i\leq3$.

Solution

The total-number cutoff gives

$$\begin{aligned} D_{\mathrm{total}} &= \binom{M+N_{\max}}{N_{\max}} \\ &= \binom{7}{3} \\ &= 35. \end{aligned}$$

The local cutoff gives

$$D_{\mathrm{local}} = (3+1)^4 = 256.$$

The local basis includes states with total occupation greater than $3$, up to total occupation $12$. The two cutoffs therefore represent different subspaces and cannot be compared by the number $3$ alone.

Cross-Links

• Many-Body and Quantum Statistical Mechanics
• Why Many-Body Physics Is Different
• Core Objects and Notation
• Thermodynamic Limit
• Occupation-Number Representation
• Tensor Products
• Tensor Products of Hilbert Spaces
• Identical Particles and Exchange Symmetry
• Fock Space and Occupation Number
• Many-Body Quantum Mechanics Crosswalk
• Sparse Matrices
• Computational Quantum Mechanics Roadmap
• Computational Many-Body Overview — how symmetry, sparsity, sign structure, and entanglement turn counting facts into method choices and qualified numerical claims.
• Exact Diagonalization Preview — how a counted finite sector becomes an indexed basis, sparse operator, eigensystem, and finite-size evidence record.
• Symmetry Sectors in Many-Body Numerics — how additive-charge filters, translation orbits, stabilizers, and compatible parity labels turn a parent count into validated numerical blocks.
• Tensor Networks Preview — structured factorizations that replace a generic exponential coefficient table when entanglement geometry is favorable.

References

• A. L. Fetter and J. D. Walecka, Quantum Theory of Many-Particle Systems, Dover (2003).
• J. W. Negele and H. Orland, Quantum Many-Particle Systems, Westview Press (1998).
• P. Coleman, Introduction to Many-Body Physics, Cambridge University Press (2015).
• R. Shankar, Principles of Quantum Mechanics, 2nd ed., Springer (1994), chapters on identical particles and many-particle systems.
• U. Schollwöck, “The density-matrix renormalization group in the age of matrix product states,” Annals of Physics 326, 96–192 (2011).
• A. Szabo and N. S. Ostlund, Modern Quantum Chemistry, Dover (1996), sections on determinant spaces and configuration interaction.