Bose-Einstein thermal operators for semidefinite optimization

Abstract: We establish that semidefinite programs (SDPs) over the unbounded positive semidefinite cone are mathematically equivalent to thermodynamic systems of independent bosonic modes: the eigenvalues of the optimization variable play the role of expected occupation numbers, the linear objective plays the role of total expected energy, and the linear equality constraints play the role of conserved non-commuting charges. Building on this perspective, we recast general SDPs as bosonic free-energy minimization problems at strictly positive temperature, regularized by the Bose-Einstein entropy; the original SDP is recovered in the zero-temperature limit. The optimal primal variable takes the form of a Bose-Einstein thermal operator parametrized by the dual variables. We prove an approximation-error bound that depends on the ground-space degeneracy and the spectral gap of the dual slack operator, improving on the linear-in-dimension worst-case duality gap of interior-point methods. We also introduce the Bose-Einstein quantum relative entropy as a Bregman divergence on the unbounded positive semidefinite cone, generated by the negative Bose-Einstein entropy. We propose it as a natural divergence for unnormalized positive operators, for which the standard Umegaki relative entropy can become negative, and we show that it satisfies a restricted monotonicity property under affine maps modeling bosonic Gaussian channels. Finally, we develop hybrid quantum-classical algorithms for the regularized SDP using only Hamiltonian simulation, Hadamard tests, and classical sampling, and bound their runtime in closed form. Unlike existing quantum SDP solvers, whose runtimes scale polynomially with an a priori upper bound on the primal trace, our framework operates directly on the unbounded cone, replacing this bound with a dependence on the spectral structure of the dual slack operator.

• The paper establishes a formal equivalence between semidefinite programming and bosonic thermal models by mapping eigenvalues to bosonic occupation numbers.
• It introduces an entropy-regularized framework that recovers the original SDP as temperature approaches zero while providing explicit, dimension-independent error bounds.
• The work develops hybrid quantum–classical algorithms using Hamiltonian simulation and Cauchy sampling to efficiently compute gradients and Hessians under favorable spectral conditions.

Bose–Einstein Thermal Operators for Semidefinite Optimization: A Technical Synthesis

Overview and Mathematical Foundations

This work establishes a formal equivalence between semidefinite programs (SDPs) over the unbounded positive semidefinite cone and the statistical mechanics of independent bosonic modes. In this correspondence, the eigenvalues of the optimization variable $X$ represent bosonic occupation numbers, the linear objective $\operatorname{Tr}[HX]$ corresponds to the expected system energy, and linear constraints $\operatorname{Tr}[Q_i X]=q_i$ are mapped to the conservation of potentially non-commuting “charges.” This perspective enables the reformulation of generic SDPs as entropy-regularized bosonic free energy minimization problems at strictly positive temperature, with a rigorous limiting procedure that recovers the original SDP as temperature approaches zero.

The free energy of the system is defined as

$$F[X] = \operatorname{Tr}[HX] - T S_{\mathrm{BE}}(X),$$

where

$$S_{\mathrm{BE}}(X) = \operatorname{Tr}\left[(X+I)\ln(X+I) - X\ln X\right]$$

serves as the Bose–Einstein entropy, characterizing occupation statistics of bosonic modes without exclusion principles or normalization requirements. The primal optimum in this regularized framework takes a closed form:

$$X_{T}(\mu) = \left(e^{(H-\mu\cdot Q)/T} - I\right)^{-1},$$

with dual variables $\mu$ functioning analogously to chemical potentials in a grand canonical ensemble. The dual objective

$$f_T(\mu) = \mu\cdot q + T\, \operatorname{Tr} \ln \left(I - e^{-\frac{1}{T}(H-\mu\cdot Q)}\right)$$

is strictly concave and amenable to first- and second-order optimization.

Entropic Regularization, Complementary Slackness, and Sharpness

A central aspect is the replacement of the unregularized SDP’s complementarity condition $K_{\mu^\star} X^\star = 0$—which enforces occupation strictly in the kernel of the dual slack operator—with a soft, temperature-parametrized generalization:

$$K_{\mu_T^\star} X_T(\mu_T^\star) = T X_T(\mu_T^\star) \ln\left[I+X_T(\mu_T^\star)^{-1}\right].$$

This smooths out the sharp occupation delineation, in analogy with thermal broadening in bosonic systems. As $\operatorname{Tr}[HX]$0, the regularized solution approaches that of the unregularized SDP, with nonzero occupation confined to the ground space of $\operatorname{Tr}[HX]$1.

Approximation Error Analysis

Critically, the authors derive explicit and dimension-independent bounds on the approximation error between the original and regularized problem:

• Entropy-based bounds apply when the feasible region is compact, scaling with the maximal Bose–Einstein entropy $\operatorname{Tr}[HX]$2. This recovers classical worst-case $\operatorname{Tr}[HX]$3 duality gap scaling.
• Spectral-gap-based bounds exhibit a notable improvement: if the slack operator $\operatorname{Tr}[HX]$4 has degeneracy $\operatorname{Tr}[HX]$5 and gap $\operatorname{Tr}[HX]$6, the error scales as $\operatorname{Tr}[HX]$7, thus, logarithmic in global dimension for fixed $\operatorname{Tr}[HX]$8, $\operatorname{Tr}[HX]$9.

Bose–Einstein Quantum Relative Entropy and Geometry

The work introduces the Bose–Einstein quantum relative entropy $\operatorname{Tr}[Q_i X]=q_i$0, a Bregman divergence generated by $\operatorname{Tr}[Q_i X]=q_i$1, suitable for the non-normalized, unbounded positive semidefinite cone where the Umegaki relative entropy can become negative. $\operatorname{Tr}[Q_i X]=q_i$2 is shown to be faithful, unitarily invariant, and additive under direct sums. Although it generally does not satisfy data-processing inequalities (DPI)—a direct consequence of its lack of joint convexity—it does obey monotonicity under physically relevant affine maps corresponding to bosonic Gaussian channels, specifically those with $\operatorname{Tr}[Q_i X]=q_i$3 for maps $\operatorname{Tr}[Q_i X]=q_i$4.

Quantum–Classical Algorithmics

The authors propose hybrid quantum–classical algorithms for evaluating gradients and Hessians of the dual problem necessary for optimization. These algorithms utilize Hamiltonian simulation, Hadamard tests, and classical Cauchy sampling, bypassing the need for quantum Gibbs sampling or block encoding. Notably, runtime bounds depend polynomially on $\operatorname{Tr}[Q_i X]=q_i$5, where $\operatorname{Tr}[Q_i X]=q_i$6 is the minimum eigenvalue of the dual slack operator, and only logarithmically on the Hilbert space dimension in the presence of a spectral gap. In contrast to prior quantum SDP solvers (e.g., those based on quantum multiplicative weights or quantum interior-point methods), no a priori upper bound on the primal trace is required.

Figure 2: The probability density function of the Cauchy distribution $\operatorname{Tr}[Q_i X]=q_i$7 for varying $\operatorname{Tr}[Q_i X]=q_i$8, illustrating the heavy-tailed sampling needed for stochastic Hamiltonian simulation in the bosonic free energy optimization framework.

Implications, Comparisons, and Open Problems

From a theoretical perspective, this work completes the “thermodynamic trilogy” of entropy-regularized SDPs (with Boltzmann, Fermi–Dirac, and now Bose–Einstein statistics, each matching a specific spectral constraint class). The Bose–Einstein regularization enables direct treatment of optimization over the full positive semidefinite cone without trace or spectral upper bounds, which is structurally aligned with quantum information tasks involving bosonic systems, continuous-variable quantum optics, and problems with unbounded resources.

Practically, these algorithms offer a route to SDP solvers respecting physically relevant geometric and information-theoretic structures and potentially exponential runtime improvement over classical IPMs in gapped regimes with low ground-space degeneracy, provided the spectral properties of the dual slack operator are favorable.

The introduction of the Bose–Einstein relative entropy as a Bregman divergence on the unbounded cone has further implications for information geometry, resource theories involving unnormalized operators, and entropy-regularized mirror descent across quantum optimization landscapes.

Open problems include:

• Characterizing the tight asymptotic relationship between the regularization temperature $\operatorname{Tr}[Q_i X]=q_i$9 and the minimum eigenvalue $$F[X] = \operatorname{Tr}[HX] - T S_{\mathrm{BE}}(X),$$0 along the optimization path to the zero-temperature limit.
• Extending efficient variance reduction techniques and mirror-descent algorithms tailored to the Bose–Einstein metric to enhance practical runtime.
• Empirical and application-driven exploration in quantum algorithms for continuous-variable systems, bosonic moment problems, and optical quantum information.

Conclusion

This paper provides a mathematically rigorous, physically grounded, and operationally significant generalization of entropy-regularized convex optimization to the unbounded semidefinite domain, via a mapping to bosonic thermodynamics. It demonstrates both structural and algorithmic advances—most strongly for cases with favorable spectral gap properties—positioning the Bose–Einstein thermal operator paradigm as a key tool for both classical and quantum algorithms addressing large-scale SDPs in quantum information, statistical physics, and convex optimization theory.