Probing disorder-induced Fisher information matrix and Cramér-Rao bound by STM

Lucas A. Oliveira, Wei Chen · 2025

Read Probing disorder-induced Fisher information matrix and Cramér-Rao bound by STM by Lucas A. Oliveira, Wei Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-22965.txt.

Introduction

Lucas Oliveira and Wei Chen bring information theory into scanning tunneling microscopy in this 2025 physics paper. A microscope of this kind maps the local density of states of a solid, point by point. Read as a probability, that density says how likely an electron at a place is to hold a given energy. In disordered solids, the density shifts from place to place. The authors build Fisher information measures for those shifts, in two mirror forms.

One form treats energy as the random variable and position as a fixed parameter. The other swaps the roles. Each form gives a Cramér-Rao bound, a floor on how much electron energy or position can vary. The authors show what these bounds mean for real microscope readings. Worked lattice models of metals and topological insulators make the ideas concrete. They present the frame as a new lens on disorder in solids.

Catalog record

CREATORLucas A. Oliveira, Wei Chen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-22965

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Probing disorder-induced Fisher information matrix and Cramér-Rao bound by STM Lucas A. Oliveira, Wei Chen Abstract The electronic local density of states of solids, if normalized correctly, represents the probability density that the electron at a specific position has a particular energy. Because this probability density can vary in space in disordered systems, we propose that one can either treat the energy as a random variable and position as an external parameter to construct a real space Fisher information matrix, or treat the position as a random variable and energy as an external parameter to construct an energy space Fisher information, both quantify the variation of local density of states caused by the disorder. The corresponding Cramér-Rao bounds in these two scenarios set a limit on the energy variance and the position variance of electrons, respectively, pointing to new interpretations of STM measurements. Our formalism thus bring the notion of information geometry into STM measurements, as demonstrated explicitly by lattice models of metals and topological insulators. I Introduction The presence of disorder can significantly modify the local and global electronic properties of solids, offering the possibility of engineering these properties by disorder, as has been practiced experimentally and commercially in semiconductors for decades [ 1 , 2 , 3 ] . Focusing on the local properties alone, various electronic properties are known to respond to the impurities in a spatially dependent manner, such as the Friedel oscillations of electron density near impurities in metals [ 4 ] . From this perspective, the invention of scanning tunneling microscope (STM) brings the experimental detection of impurity effects to a whole new level, since it enables the detection of the spatial and energy dependence of electronic local density of states (LDOS) down to atomic scale [ 5 , 6 , 7 , 8 , 9 ] . From this LDOS, various impurity effects have been experimentally verified, such as the impurity bound states [ 10 , 11 ] and quasiparticle interference [ 12 ] in cuprate superconductors, among countless many others, offering a direct experimental confirmation of various theoretical predictions. In this paper, we connect the disorder-induced LDOS to an arena that has not been explored much in condensed matter physics, but has been heavily exploited in statistics, namely the information geometry. We observe that if one combines the spatial and energy dependence of LDOS with the Born interpretation of quantum mechanical wave function, which states that correctly normalized LDOS \rho({\bf x},E) is the probability density of finding an electron at a specific position {\bf x}=(x^{1},x^{2}...x^{D}) in a D -dimensional lattice with a particular energy E , then the notion of classical Fisher information matrix (FIM) naturally arise [ 13 ] . Statistically, the notion of FIM arises when one has a probability distribution over some random variable whose profile depends on certain external parameters, where the FIM characterizes the variation of the probability distribution in the parameter space. Because of the probability interpretation of LDOS, one can adopt two scenarios to construct an FIM from the standard procedure in the probability theory. In the first scenario, one treats the energy E as a random variable and the position {\bf x} as an external parameter, i.e., the normalized LDOS \rho({\bf x},E) is regarded as the probability of finding an electron within energy window E and E+\Delta E at a specified position {\bf x} , which allows to introduce a real space FIM to describe the variation of LDOS in real space caused by the disorder. In the second scenario, one reverses the roles by treating the position {\bf x} as a random variable and the energy E as an external parameter, meaning that the normalized LDOS \rho({\bf x},E) is regarded as the probability of finding an electron at position {\bf x} with a specified energy E , which amounts to an energy space FIM that also quantifies the variation of LDOS. The real space FIM and the energy space FIM, respectively, describe the distortion of the D -dimensional real space and the one-dimensional energy space caused by the disorder, thereby bringing the notion of information geometry [ 14 , 15 ] into condensed matter physics. Furthermore, note that FIM has been applied to various branches of condensed matter physics, such as density functional theory [ 16 ] , phase transitions [ 17 ] , diamagnetism [ 18 ] , etc. Our formalism thus demonstrates the LDOS as another application of FIM in a way that is completely measurable by STM. In statistics, an important consequence of the FIM is the Cramér-Rao bound (CRB), which sets a limit on the accuracy of parameter estimation [ 19 , 20 ] . For the real space FIM, we will elaborate that the corresponding CRB has a remarkable interpretation in electronic systems, namely it dictates that the variance of local energy \langle E^{2}\rangle-\langle E\rangle^{2} must be greater than a product of the spatial derivative of the local energy \partial_{\mu}\langle E\rangle and the inverse of the FIM, regardless the detail of the host material and what kind of disorder is present. In comparison, the energy space FIM gives a CRB that limits the spatial variance of electron position \langle(x^{\mu})\rangle^{2}-\langle x^{\mu}\rangle^{2} according to the energy derivative of the average position \partial_{E}\langle x^{\mu}\rangle and the FIM, also applicable to any kind of material and disorder. These bonuds on the energy variance and position variance of electrons point to new interpretations and data analysis of LDOS that can be broadly applied to any STM measurements, as will be demonstrated explicitly using theoretical models of metals and topological insulators. II Real space Fisher information matrix and Cramér-Rao bound II.1 Real space Fisher information matrix defined from local density of states We consider a D -dimensional lattice with unit vectors {\bf a}^{\mu}=a^{\mu}{\hat{\bm{\mu}}} , where \mu=1\sim D is reserved for the spatial directions. The locations of the unit cell, or lattice sites, are described by the Bravais lattice vector {\bf x}=\sum_{\mu=1}^{D}n_{\mu}{\bf a}^{\mu} , where n_{\mu} are integers. Each unit cell may have several species of electrons labeled by \sigma , which may signify spin, orbit, sublattice, etc. Generically, the lattice Hamiltonian takes the form \displaystyle H=\sum_{\bf xx^{\prime}\sigma\sigma^{\prime}}t_{\bf xx^{\prime}\sigma\sigma^{\prime}}|{\bf x},\sigma\rangle\langle{\bf x}^{\prime},\sigma^{\prime}|, (1) where |{\bf x},\sigma\rangle is the ket state of an electron of species \sigma at site {\bf x} . The lattice eigenstates |E_{\ell}\rangle and eigenenergies E_{\ell} are found through diagonalizing the Hamiltonian H|E_{\ell}\rangle=E_{\ell}|E_{\ell}\rangle . The wave function of species \sigma at site {\bf x} of an eigenstate |E_{\ell}\rangle is denoted by \displaystyle\langle{\bf x},\sigma|E_{\ell}\rangle=\tilde{a}_{\ell\sigma}({\bf x}). (2) Throughout the article, we reserve the subscript n for the filled states at zero temperature within the energy range of an STM measurement E_{c}<E_{n}<0 , where E_{c} denotes the lowest energy measured by STM (typically few eV), and likewisely for the summation \sum_{n}\equiv\sum_{E_{c}<E_{n}<0} . Certainly STM also measures positive energy states E_{\ell}>0 up to some highest energy, but they are unimportant for our discussion since the electrons at zero temperature do not occupy these states. Furthermore, usually STM do not measure all the degrees of freedoms \sigma within a unit cell separately. For instance, the two spins are usually not resolved by STM. We will assume that the STM can resolve the LDOS on each sublattice \gamma within a

Read the complete text