Bose–Einstein Thermal Operator

• Bose–Einstein thermal operator is defined via a bosonic occupation law and serves as the unique optimizer in a free‐energy regularization of semidefinite programs.
• It links the spectral properties of the dual slack operator—specifically the ground-space degeneracy and spectral gap—to controlled approximation guarantees in optimization.
• The operator also appears in finite-temperature Bose–Einstein condensate theory, enabling phase-space sampling through a thermal density formulation for robust simulation.

The Bose–Einstein thermal operator is an operator-valued Bose occupation law that arises when a semidefinite program over the unbounded positive semidefinite cone is recast as a bosonic free-energy minimization problem at strictly positive temperature. In that formulation, the primal variable is the unique stationary and optimal operator

$X(\beta,y)=\bigl(e^{\beta\,S(y)}-I\bigr)^{-1},$

with $S(y)=C-\sum_{i=1}^m y_iA_i$ the dual slack operator and $\beta=1/T$ the inverse temperature; the original semidefinite program is recovered in the zero-temperature limit (Minervini et al., 26 May 2026). A related but distinct use of the term appears in finite-temperature Bose–Einstein condensate theory, where the thermal density operator is written as $\hat\rho=\exp[-\beta \hat K]/Z$ for a generalized grand-canonical generator $\hat K$ that includes a nonlinear chemical-potential term, enabling phase-space sampling in Wigner and positive-$P$ representations (Ng et al., 2018). Across these settings, the phrase denotes a thermodynamic encoding of bosonic occupation structure, but the optimization-theoretic and many-body constructions serve different purposes.

1. Semidefinite-programming formulation

A standard primal semidefinite program in cost–constraint form is

$$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$

Its Lagrangian dual introduces multipliers $y\in\mathbb R^m$ and the dual slack operator

$S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$

constrained by $S(y)\succeq0$ (Minervini et al., 26 May 2026).

The thermodynamic interpretation identifies the eigenvalues of the optimization variable with expected occupation numbers, the linear objective with total expected energy, and the linear equality constraints with conserved non-commuting charges (Minervini et al., 26 May 2026). In that sense, the SDP is mathematically equivalent to a thermodynamic system of independent bosonic modes. This equivalence is not merely heuristic: it is used to replace the sharp optimization problem by a free-energy minimization at strictly positive temperature.

The temperature parameter is $S(y)=C-\sum_{i=1}^m y_iA_i$0, with inverse temperature or hardness parameter

$S(y)=C-\sum_{i=1}^m y_iA_i$1

At fixed $S(y)=C-\sum_{i=1}^m y_iA_i$2, the bosonic free-energy regularization smooths the optimization over the unbounded cone while preserving the original constraint structure through the dual variables.

2. Free energy, entropy, and the thermal operator

The regularized free-energy problem is

$S(y)=C-\sum_{i=1}^m y_iA_i$3

where the Bose–Einstein operator entropy is defined from the scalar bosonic entropy function

$S(y)=C-\sum_{i=1}^m y_iA_i$4

by the matrix functional

$S(y)=C-\sum_{i=1}^m y_iA_i$5

The corresponding primal free energy is

$S(y)=C-\sum_{i=1}^m y_iA_i$6

and one solves

$S(y)=C-\sum_{i=1}^m y_iA_i$7

to recover the original SDP as $S(y)=C-\sum_{i=1}^m y_iA_i$8 (Minervini et al., 26 May 2026).

Taking the derivative of the Lagrangian with respect to $S(y)=C-\sum_{i=1}^m y_iA_i$9 and setting it to zero yields

$\beta=1/T$0

Accordingly, the optimal primal variable at dual parameter $\beta=1/T$1 is the Bose–Einstein thermal operator

$\beta=1/T$2

The operator is therefore the bosonic occupation function applied to the dual slack spectrum. In the optimization setting, this is the central object: it is the unique stationary and hence optimal primal operator (Minervini et al., 26 May 2026).

A plausible implication is that the regularization is adapted to the geometry of the unbounded positive semidefinite cone in a way that standard trace-bounded formulations are not, because the bosonic entropy naturally accommodates unnormalized positive operators.

3. Zero-temperature limit and spectral structure

As $\beta=1/T$3, the scalar Bose occupation law

$\beta=1/T$4

for eigenvalues $\beta=1/T$5 of $\beta=1/T$6, concentrates support on the zero-eigenspace of the dual slack operator (Minervini et al., 26 May 2026). In this limit, the support of $\beta=1/T$7 collapses onto that zero-eigenspace, enforcing complementary slackness $\beta=1/T$8 and recovering the sharp SDP solution.

The spectral analysis is expressed in terms of the optimal dual slack $\beta=1/T$9, whose spectrum is written as

$\hat\rho=\exp[-\beta \hat K]/Z$0

Here $\hat\rho=\exp[-\beta \hat K]/Z$1 is the ground-space degeneracy and $\hat\rho=\exp[-\beta \hat K]/Z$2 is the spectral gap above the ground space (Minervini et al., 26 May 2026).

The approximation error obeys

$\hat\rho=\exp[-\beta \hat K]/Z$3

From this, a choice

$\hat\rho=\exp[-\beta \hat K]/Z$4

together with an exponentially mild correction in $\hat\rho=\exp[-\beta \hat K]/Z$5 suffices to reduce the SDP duality gap below any $\hat\rho=\exp[-\beta \hat K]/Z$6 (Minervini et al., 26 May 2026). The paper states that this improves on the linear-in-dimension worst-case duality gap of interior-point methods. It also states that when $\hat\rho=\exp[-\beta \hat K]/Z$7 and $\hat\rho=\exp[-\beta \hat K]/Z$8 remain small, for example polylogarithmic in the dimension, the required $\hat\rho=\exp[-\beta \hat K]/Z$9 grows only polylogarithmically rather than linearly in $\hat K$0.

This spectral-gap dependence is one of the main distinguishing features of the framework. A common misconception would be to treat the regularization error as controlled only by ambient dimension; the stated bound shows that the relevant quantities are instead the ground-space degeneracy and the spectral gap of the dual slack operator (Minervini et al., 26 May 2026).

4. Relative entropy on the unbounded cone

The Bose–Einstein quantum relative entropy is introduced as the matrix Bregman divergence generated by $\hat K$1: $\hat K$2 Equivalently,

$\hat K$3

This divergence is proposed as a natural divergence for unnormalized positive operators, for which the standard Umegaki relative entropy can become negative (Minervini et al., 26 May 2026).

Its basic properties are explicitly stated. It satisfies faithfulness, $\hat K$4, with equality if and only if $\hat K$5; unitary invariance; additivity under direct sums; and strict convexity in the first argument (Minervini et al., 26 May 2026). At the same time, it fails to be jointly convex, and hence there is no general CPTP data-processing inequality.

The absence of general CPTP monotonicity is an important limitation rather than a defect hidden by the formulation. The paper instead proves a restricted monotonicity statement for affine maps modeling bosonic Gaussian channels,

$\hat K$6

under which

$\hat K$7

The stated examples are lossy channels $\hat K$8, amplifiers $\hat K$9, and additive-noise channels $P$0 (Minervini et al., 26 May 2026).

This suggests that the divergence is tailored to bosonic operator geometry rather than to the full CPTP framework ordinarily associated with normalized density operators.

5. Hybrid quantum–classical algorithms

The optimization framework is accompanied by hybrid quantum–classical algorithms for the regularized SDP (Minervini et al., 26 May 2026). A key primitive is an unbiased estimator of $P$1 for any Hermitian $P$2.

The starting point is the series expansion

$P$3

This series is truncated at

$P$4

and each term $P$5 is represented as

$P$6

Sampling $P$7 from the Cauchy law and using the Hadamard test on the unitary $P$8 yields a Monte Carlo estimate of each term (Minervini et al., 26 May 2026).

The resource statement given in the paper is that one combines $P$9 such shots to reach precision $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$0, and Hamiltonian-simulation techniques implement $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$1 with cost linear in $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$2 and the $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$3-norm of the decomposition of $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$4 (Minervini et al., 26 May 2026). A similar double-series plus controlled-SWAP circuit estimates Hessian matrix elements $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$5.

The abstract emphasizes the contrast with existing quantum SDP solvers: unlike runtimes that scale polynomially with an a priori upper bound on the primal trace, this framework operates directly on the unbounded cone, replacing that bound with a dependence on the spectral structure of the dual slack operator (Minervini et al., 26 May 2026). This does not remove spectral assumptions; rather, it changes the controlling parameter from a trace bound to slack-spectrum data.

6. Finite-temperature Bose–Einstein condensates and phase-space realizations

In finite-temperature Bose–Einstein condensate theory, a distinct but related construction begins from a generalized grand-canonical generator

$$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$6

where $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$7 is the Bose Hamiltonian, $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$8 the total number operator, and $$\min_{X\succeq0}\;\bigl\langle C,X\bigr\rangle \quad\text{subject to}\quad \langle A_i,X\rangle = b_i\quad (i=1,\dots,m)\,.$$9 is a generalized chemical potential Taylor-expanded to second order (Ng et al., 2018). The thermal density operator is then

$y\in\mathbb R^m$0

In practice, $y\in\mathbb R^m$1 and $y\in\mathbb R^m$2 are chosen so as to cancel unwanted linear and quadratic terms in the Bogoliubov expansion (Ng et al., 2018).

After diagonalizing $y\in\mathbb R^m$3 to quadratic order in fluctuations, one obtains a product of independent bosonic modes $y\in\mathbb R^m$4. In the diagonal basis, each mode $y\in\mathbb R^m$5 is a thermal Bose state with

$y\in\mathbb R^m$6

plus $y\in\mathbb R^m$7-quantum of Wigner noise in the Wigner representation (Ng et al., 2018). The initial Wigner field is

$y\in\mathbb R^m$8

with Gaussian random variables satisfying

$y\in\mathbb R^m$9

For the positive-$S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$0 representation, the density matrix is expanded as

$S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$1

and Gaussian sampling is carried out through the normal-ordered covariance matrix $S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$2 and a square root $S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$3 satisfying $S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$4 (Ng et al., 2018).

A central technical point is the zero-momentum regularization. In the usual Bogoliubov linearization, a quadratic term proportional to $S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$5 diverges in the $S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$6 channel if only a linear $S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$7 is used. Including $S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$8 yields

$S(y)\;=\;C\;-\;\sum_{i=1}^m y_i A_i\,,$9

with $S(y)\succeq0$0 and $S(y)\succeq0$1, so the choice

$S(y)\succeq0$2

exactly cancels the $S(y)\succeq0$3 term, removing the zero-momentum phase-diffusion divergence and rendering the quadratic form diagonalizable including $S(y)\succeq0$4 (Ng et al., 2018).

With this choice, one obtains the diagonal approximate Hamiltonian

$S(y)\succeq0$5

and the mode expansion

$S(y)\succeq0$6

In the homogeneous example $S(y)\succeq0$7, $S(y)\succeq0$8 finite, the familiar plane-wave solutions are recovered: $S(y)\succeq0$9

$S(y)=C-\sum_{i=1}^m y_iA_i$00

and for $S(y)=C-\sum_{i=1}^m y_iA_i$01 the regularization gives $S(y)=C-\sum_{i=1}^m y_iA_i$02, $S(y)=C-\sum_{i=1}^m y_iA_i$03, $S(y)=C-\sum_{i=1}^m y_iA_i$04 (Ng et al., 2018).

The practical connection is explicit: one samples Gaussian phonon occupations, together with Wigner or normal-ordered positive-$S(y)=C-\sum_{i=1}^m y_iA_i$05 noise, and back-transforms to obtain initial $S(y)=C-\sum_{i=1}^m y_iA_i$06-number fields for subsequent stochastic evolution (Ng et al., 2018). This use of a thermal operator is conceptually related to the optimization-theoretic Bose–Einstein operator through the same bosonic occupation law, but the objects differ: one is a primal optimizer on the unbounded positive semidefinite cone, the other a thermal density operator for finite-temperature condensate states.

7. Conceptual scope and limitations

The optimization construction and the condensate construction share a common statistical-mechanical vocabulary, but they are not interchangeable. In semidefinite optimization, the Bose–Einstein thermal operator is

$S(y)=C-\sum_{i=1}^m y_iA_i$07

the unique minimizer of a bosonic free-energy functional, and its principal significance is the recovery of the SDP in the zero-temperature limit together with spectral-gap-based approximation guarantees (Minervini et al., 26 May 2026). In finite-temperature condensate theory, the thermal operator is instead the grand-canonical density operator $S(y)=C-\sum_{i=1}^m y_iA_i$08, used to generate initial conditions for Wigner or positive-$S(y)=C-\sum_{i=1}^m y_iA_i$09 phase-space simulations (Ng et al., 2018).

Two limitations are stated explicitly in the source material. First, the Bose–Einstein relative entropy fails to be jointly convex and therefore does not furnish general CPTP data processing (Minervini et al., 26 May 2026). Second, the phase-space construction for condensates relies on a nonlinear chemical potential, specifically the choice $S(y)=C-\sum_{i=1}^m y_iA_i$10, to remove the $S(y)=C-\sum_{i=1}^m y_iA_i$11 phase-noise divergence and obtain a diagonalizable quadratic theory (Ng et al., 2018). These are not peripheral details; they delimit the regimes in which the corresponding thermal-operator formalisms are mathematically well behaved.

A plausible implication is that the phrase “Bose–Einstein thermal operator” denotes a family resemblance rather than a single universal object: in each case, bosonic occupation statistics define the operator structure, but the surrounding mathematical role is determined by whether the problem is an SDP over the unbounded cone or a finite-temperature field theory represented in phase space.