Thermal balance and the amplifier entropy photon-number inequality
Harjas Sandhu
Abstract
Quantum amplification adds unavoidable noise. A variational argument seeks the sharp lower bound on output entropy for two independent inputs, even when the modes within either input are correlated. The result. The entropy photon number expresses a state’s entropy as the occupation of a thermal state with the same entropy per mode. The manuscript presents a variational argument for its sharp amplifier lower bound: at gain G, the output entropy photon number is at least G times the first input’s value plus (G − 1) times the sum of the second input’s value and one. Thermal inputs attain the bound. For example, gain 2 and thermal occupations 1 and 0 give output occupation 3. The proposed inequality says that independent inputs with those same entropies cannot produce a smaller output entropy photon number. How the argument works. The amplifier’s second input enters through a creation operator. That reversal changes the sign structure of a logarithmic-mean comparison and introduces the extra quantum-noise term. Thermal relaxation obeys a matching signed balance. At a regularized entropy minimum, the two structures cancel the energy corrections. A thermal Stein estimate is then used to turn small balance defects into trace-norm closeness to thermal states, with finite-energy estimates closing the proposed argument. The inequality was conjectured by De Palma, Mari and Giovannetti. The variational framework and four-weight interpolation adapt OpenAI’s beam-splitter argument. Scope. The general argument concerns two independent finite-energy input groups with any fixed finite number of modes. Correlations and entanglement within either group are allowed. Correlations between the two groups can violate the independent-input formula. The manuscript also studies Gaussian spectral refinements and Gaussian equality. General non-Gaussian equality and a quantitative stability theorem remain open. The proposed general amplifier argument and its Gaussian special cases should be read at their respective scopes.