arXiv2018
It is "conventional wisdom" that the uncertainty of local temperature measurements on equilibrium systems diverges exponentially fast as their temperature $T$ drops to zero. In contrast, some exactly solvable models showcase a more benign power-law-like scaling, when only a small non-equilibrium fragment of the equilibrium system is measured. Does this mean that a part may contain more information about the global temperature than the whole? Certainly not. Here, we resolve this apparent paradox. First, we prove that local quantum thermometry at low $T$ is exponentially inefficient in non-critical, gapped, and infinite spin and harmonic lattices. In contrast, we show through an open-system analysis, that the thermal sensitivity of a harmonic thermometer (probe) jointly equilibrated with a reservoir (sample) by means of an Ohmic coupling scheme, displays a distinctive power-law-like behavior as $T \rightarrow 0$. To reconcile these two results, we exploit the fact that local thermometry on a harmonic chain may be viewed as a temperature measurement on an oscillator in a discretized harmonic environment: A gapped translationally invariant chain (for which low--$T$ thermometry is indeed exponentially inefficient) maps into a non-standard open-system model where the low-frequency modes of the sample decouple from the probe. On the contrary, a gapless instance of such chain gives rise to the canonical Ohmic probe-sample interaction, which does include sample modes of arbitrarily low frequency. In this way, we show that the power-law-like thermometric performance observed in typical dissipative models stems from the fact that these are gapless and, as such, not subjected to the exponential limitations of their gapped counterparts. The key feature of many-body systems, when it comes to the ultimate limitations of low--$T$ thermometry, is thus whether or not their energy spectrum is gapped.