---
title: Maximally Mixed Invariant State (MMIS)
url: https://www.emergentmind.com/topics/maximally-mixed-invariant-state-mmis
type: topic
---

# Maximally Mixed Invariant State (MMIS)

A maximally mixed invariant state (MMIS) is a mixed quantum state constructed by projecting onto the subspace of a many-body Hilbert space invariant under a specified symmetry group, or, more generally, by projecting onto the ground space or an invariant sector of a local Hamiltonian and normalizing. The MMIS exhibits distinct structural, entropic, and correlation properties compared to generic mixed states, including entanglement, area law bounds, and tensor network approximability. MMISs arise both as unique steady states of symmetric quantum channels and as canonical mixed states in the study of quantum many-body entanglement under symmetry or local Hamiltonian constraints.

## 1. Definitions and Classes of MMIS

Let $H = \sum_i h_i$ be a local Hamiltonian on $n$ sites (each site of local Hilbert space dimension $d$) with ground space projector $\Pi_{\text{gs}}$ and ground-state degeneracy $r = \dim \mathrm{Eig}(H,0)$. The maximally mixed ground state is
$$
\Omega \equiv \frac{\Pi_{\text{gs}}}{r}.
$$
$\Omega$ is invariant under $H$ and all $h_i$, satisfying $[H, \Omega]=0$ and $H \Omega=0$. As $\beta \to \infty$, $\Omega$ is the zero-temperature limit of the Gibbs state, i.e., $\Omega = \lim_{\beta \rightarrow \infty} e^{-\beta H}/Z_\beta$, with $Z_\beta = \mathrm{Tr}\,e^{-\beta H}$ [2310.19028].

For a compact symmetry group $G$ acting onsite, let $V^{\otimes N}$ denote $N$ qudits each carrying a $d$-dimensional irreducible representation (irrep) $V$ of $G$. The total Hilbert space decomposes as $V^{\otimes N} = \bigoplus_J C_J^N V_J$, with $J$ labelling irreps, $d_J$ their dimensions, and $C_J^N$ their multiplicities. Projecting onto the $G$-invariant (trivial irrep) sector yields [2406.08542]
$$
\rho_{\text{inv}} = \frac{P_{\text{inv}}}{\operatorname{Tr} P_{\text{inv}}},
$$
where $P_{\text{inv}}$ is the projector onto $(V^{\otimes N})^G$. Analogous definitions hold for MMISs invariant under translation (using the projector onto the translation-invariant subspace) [2605.15201].

For two-qubit systems, a density matrix is "locally maximally mixed" (LMM) if each single-qubit reduced density matrix is maximally mixed: $\operatorname{Tr}_{2}\rho = \frac{1}{2}I$, $\operatorname{Tr}_{1}\rho = \frac{1}{2}I$ [2304.13555]. For $N$-party systems, a $k$-uniform mixed state is one for which every reduction to any subset of $k$ parties is maximally mixed [1906.01311].

## 2. Entanglement and Correlation Properties

MMISs depart from the intuition that mixedness equals classicality: under symmetries or as ground space projectors, they can display strong nonclassical and entangled features.

**Exact bipartite entanglement in symmetry sectors (group-invariant MMIS):** For non-Abelian $G$, the entanglement of formation ($E_F$) and distillation ($E_D$) of $\rho_{\text{inv}}$ are exactly computable and coincide for any bipartition. For region $A$ versus $A^c$,
$$
E_F(\rho_{\text{inv}}) = E_D(\rho_{\text{inv}}) = \sum_J p_J \log d_J,
$$
where $p_J$ is proportional to $C_J^{N_A} C_{J^*}^{N_B}$, the multiplicities of $J$ and its conjugate $J^*$ in $A$ and $A^c$ [2406.08542]. For Abelian $G$ (all $d_J=1$), MMIS is separable; for non-Abelian $G$, it is entangled. For compact semisimple Lie groups, $E(\rho_{\text{inv}}) \sim (\dim \mathfrak{g} - \operatorname{rank} \mathfrak{g})/4 \log N$ for a bipartition $N_A:N_B$ [2406.08542].

**Long-range entanglement from dimensional constraints (translation-invariant MMIS):** For the translation-invariant MMIS $\rho_T$ on a 1D ring,
- The dimension of the translation-invariant subspace grows as $\sim d^L/L$.
- SRE (short-range entangled) decompositions can only span $\mathrm{poly}(L)$ states, so $\rho_T$ cannot be expressed as a convex combination of SRE pure states.
- The conditional mutual information $I(A:C|B)$ scales as $\log L$ in the $L \to \infty$ limit for partitions $A,B,C$ with $|A|,|B|,|C| \sim L$ [2605.15201].
- Rényi-$\alpha$ operator-space entanglement satisfies $S_{\alpha>1} \sim \frac{\alpha}{\alpha-1}\log L$, while $S_1$ scales with the volume of the region.

**Local indistinguishability:** For fixed $k \ll N$, any $k$-site reduced density matrix of $\rho_{\text{inv}}$ converges to the local maximally mixed state in the $N \to \infty$ limit (errors $O(1/N)$ for Lie groups), despite global long-range entanglement [2406.08542].

**Higher-order k-uniform MMIS and nonclassicality:** A $k$-uniform mixed state has all $k$-party marginals maximally mixed, generating multipartite entanglement and, for $k\ge2$, vanishing all lower-order correlation tensors. Optimal constructions via commuting Pauli generators or orthogonal arrays can achieve the maximal known purity for given $N,k$; these states display genuine multipartite entanglement and Bell inequality violation [1906.01311].

**Summary of correlation and entanglement scaling:**

| MMIS Class         | Entanglement Scaling          | Local Reductions             | Operator-Space Entropy      |
|--------------------|------------------------------|------------------------------|-----------------------------|
| Symmetry sector    | $E \sim c \log N$ [non-Abelian] | Maximally mixed for $k \ll N$ | $S_1\sim$ volume, $S_{\alpha>1}\sim \log N$ |
| Translation MMIS   | $E_f \leq \log L$               | Maximally mixed for $|A|\ll L$ | $S_{\alpha>1}\sim \log L$             |
| $k$-uniform MMIS   | See [1906.01311], maximal GME  | All $k$-party marginals maximally mixed | -                   |

## 3. Area Laws and Bounds on Mutual Information

In gapped 1D local Hamiltonians ($d=O(1)$, constant spectral gap $\gamma$), for MMIS $\Omega=\Pi_{\text{gs}}/r$:

- For contiguous $L\subset\{1,\ldots,n\}$ and $\varepsilon>0$, the $\varepsilon$-smoothed max-mutual information obeys
  $$
  I_{\max}^\varepsilon(L:L^c)_\Omega \leq O(\log |L| + \log(1/\varepsilon)).
  $$
- The ordinary mutual information satisfies
  $$
  I(L:L^c)_\Omega \leq O(\log |L|).
  $$
These bounds are independent of the ground state degeneracy and rely on the construction of a "good" approximate ground-state projector (AGSP), i.e., an operator $K$ with controlled growth of Schmidt rank and decay outside the ground space [2310.19028].

For 2D frustration-free, locally gapped Hamiltonians, with good AGSPs whose Schmidt rank grows as $\exp(O(\operatorname{polylog} |\partial L|))$, one obtains
$$
I_{\max}^\varepsilon(L:L^c)_\Omega \leq O(|\partial L| \cdot \operatorname{polylog} |L| + \log(1/\varepsilon)),
$$
thus demonstrating a polylogarithmic correction to a strict area law [2310.19028].

## 4. Tensor Network and MPO Approximations

For any fixed $\varepsilon>0$, there exists a Hermitian Matrix Product Operator (MPO) $M$ with bond dimension $\mathrm{poly}(n/\varepsilon)$ such that
$$
\|\Omega - M/\operatorname{Tr}M\|_1 \leq \varepsilon,
$$
and $\operatorname{Tr}M$ approximates $r$, the dimension of the ground space, to small relative error. The construction uses an AGSP written as a low-degree polynomial filter (Chebyshev-type) in $H$, implemented as an MPO with bond dimension $\mathrm{poly}(\deg)$ [2310.19028]. For 1D gapped Hamiltonians, this guarantees a tensor network description of the MMIS scalable to large sizes and arbitrary trace-norm accuracy.

## 5. Invariant Theory and Classification in Low-Dimensional Cases

For two-qubit "locally maximally mixed" states, the complete set of rational local unitary invariants is given by the three elementary polynomial invariants:
- $t_2 = \mathrm{Tr}(CC^T)$
- $t_3 = \det C$
- $t_4 = \mathrm{Tr}((CC^T)^2)$

Here $C$ is the $3 \times 3$ real correlation matrix in the Bloch representation. The field of invariants is purely transcendental $\mathbb{R}(t_2,t_3,t_4)$, and the geometric quotient under LU is $\mathbb{R}^3_{(t_2,t_3,t_4)}$ [2304.13555]. Orbit classification is thereby reduced to computing this triple, with explicit correspondence for families such as Werner states and Bell-state mixtures.

## 6. Mixed-State Phase Signatures and Spontaneous Symmetry Breaking

MMISs in symmetry-protected or symmetry-enforced sectors can exhibit "strong-to-weak" spontaneous symmetry breaking (SW-SSB), in which the ensemble of pure-state components displays broken "strong" product symmetry (e.g., $T \otimes \overline{T}$ for translations) but only the "weak" diagonal is preserved in the whole MMIS. Nonlinear correlators such as variance-normalized Rényi-2 order parameters remain $O(1)$ at large size, reflecting global order invisible to local probes. Meanwhile, all linear correlators vanish exponentially or algebraically, confirming the absence of conventional long-range order [2605.15201][2406.08542].

## 7. Preparation and Physical Realization

MMISs naturally arise as unique steady states of strongly symmetric quantum channels. For instance, the translation-invariant MMIS on a ring is stabilized by a Lindblad master equation employing a collective jump operator that commutes with translations; numerically, relaxation to the MMIS occurs at an exponential rate [2605.15201]. In the context of Hamiltonian ground spaces, cold-atom or trapped-ion platforms with configurable local terms and global symmetries offer routes to physical preparation of such states.

---

MMISs unify and generalize several strands in the quantum many-body literature: as ground space projectors, group-invariant mixed states, translation-invariant maximally mixed states, and $k$-uniform mixed states, they allow precise entanglement calculation, admit efficient tensor network representation, and display robust forms of long-range entanglement and mixed-state ordering determined by underlying symmetries and dimension constraints [2310.19028][2406.08542][2605.15201][2304.13555][1906.01311].

Source: https://www.emergentmind.com/topics/maximally-mixed-invariant-state-mmis