Weak Convexity of Fisher Information Matrix and Superresolved Localization of Blinking Sources of Light

Dmitri B. Horoshko, Alexander B. Mikhalychev, Fedor Jelezko, Polina P. Kuzhir · 2025

Read Weak Convexity of Fisher Information Matrix and Superresolved Localization of Blinking Sources of Light by Dmitri B. Horoshko, Alexander B. Mikhalychev, Fedor Jelezko, Polina P. Kuzhir (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-2501-14674.txt.

Introduction

In a 2025 arXiv paper, Horoshko and co-authors ground blinking light tricks in quantum measure theory. Single-molecule scope methods place light dots with nanometre care by making each dot blink. One bright dot at a time can be placed well past the classic blur cap. The team asks how much place data the light truly holds.

That cap is set by Fisher information, a score of how much a read can tell. Blinking pays off, they prove, because that score bulges in a set way when states mix. The edge sits in the light itself, before any read takes place.

Catalog record

CREATORDmitri B. Horoshko, Alexander B. Mikhalychev, Fedor Jelezko, Polina P. Kuzhir
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE80 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-14674

Browse the full library · About the Tower of Babel

Full text

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

Weak Convexity of Fisher Information Matrix and Superresolved Localization of Blinking Sources of Light Dmitri B. Horoshko, Alexander B. Mikhalychev, Fedor Jelezko, Polina P. Kuzhir Abstract A group of techniques known by the general name of single-molecule localization microscopy reaches a nanometer-scale spatial resolution of point light emitters, well below the diffraction limit of the traditional microscopy. The key feature of these techniques is blinking, alternation of bright and dark states, of each emitter so that no more than one emitter is bright within the width of the point-spread function of the microscope during a time sufficient for its localization. We give a formulation of the optical part of these techniques in terms of quantum metrology, where the limit of precision is determined by the Fisher information on the emitters positions contained in the measurement data. We show that the advantage in resolution provided by making the emitters blink is a consequence of the fundamental property of Fisher information, its convexity. In particular, we prove the weak matrix convexity and the trace convexity of the Fisher information matrix – two fundamental results in the multiparameter estimation theory. We show also that the advantage in information is inherent to the quantum state of light itself before the measurement and is related to the convexity of the quantum Fisher information matrix. I Introduction The lateral resolution of a conventional far-field microscope is fundamentally restricted by the diffraction limit, equal to about half the wavelength of the illuminating light [ 1 ] , which is above 200 nm in the visible range. In the last 30 years, numerous methods for surpassing the diffraction limit have been developed and are known by the general name of superresolution microscopy [ 2 , 3 , 4 ] . One of the most successful directions in this area of research is related to subdiffraction imaging of point-like incoherent sources of light represented by fluorescent molecules [ 5 , 6 , 7 ] , achieving nanometer-scale lateral resolution and distinguished by the Nobel Prize in Chemistry in 2014 ’for the development of superresolved fluorescence microscopy.’ An important subdirection is represented by the techniques of stochastically activated (blinking) fluorophores such as photoactivated localization microscopy (PALM) [ 8 ] , stochastic optical reconstruction microscopy (STORM) [ 9 ] , points accumulation for imaging in nanoscale topography (PAINT) [ 10 ] , and their variants, know by the general name of single-molecule localization microscopy (SMLM) and having found numerous applications in cell biology [ 6 ] . The SMLM techniques have also been applied to color centers in diamond providing subdiffraction-limit sensing of magnetic or electric fields [ 11 , 12 ] . The general principle of SMLM consists in making the fluorophores blink, by a mechanism specific to each technique, so that not more than one source is present within the width of the point-spread function (PSF) of the microscope during a time sufficient for its precise localization. The latter is achieved by a simple arithmetic mean of the photodetection positions or by taking an analytical model for PSF and estimating the position of its peak by least squares or maximum likelihood estimation [ 13 ] . The standard deviation of the source position scales as 1/\sqrt{N} with the number N of detected photons [ 14 ] , which allowed one to localize isolated fluorophores with an uncertainty of about 1 nm by detecting N\approx 10^{4} photons for each of them by a technique known as fluorescence imaging with one-nanometer accuracy (FIONA) [ 15 ] . Similarly, the positions of multiple sources within PSF can be found by modeling the intensity distribution on the microscope camera by a sum of PSFs and estimating the positions of its peaks, albeit with a much larger uncertainty [ 16 , 17 ] . The 1/\sqrt{N} scaling of the position standard deviation is easily understood in the framework of mathematical statistics, namely, its branch known as point estimation theory [ 18 , 19 ] . The central relation of this theory is the Cramér-Rao bound for the variance of any unbiased estimator of the parameter x : \left(\Delta x\right)^{2}\geq 1/\underline{F}, (1) where \underline{F} is the Fisher information on this parameter contained in the measurement data. Here and below underlined letters denote quantities related to the entire measurement session typically consisting of many photodetections. In the case of localization of a single emitter, the subsequent photodetections are statistically independent, and the total Fisher information carried by N photons is simply \underline{F}=NF_{\text{1phot}} , where F_{\text{1phot}} is the Fisher information on the emitter position carried by one photon. The Cramér-Rao bound can be asymptotically reached in the high N limit when the estimator is given by the maximum likelihood estimation [ 18 , 19 ] . Thus, the 1/\sqrt{N} scaling of the position standard deviation is a mere consequence of the additivity of Fisher information with respect to independent observations. The single-photon Fisher information can be calculated for various models of the PSF and the accompanying noise [ 20 , 21 ] providing the Cramér-Rao bound for any given N . The variance of the estimated emitter position was shown to approach the Cramér-Rao bound for N=10^{2}-10^{4} photons allowing thus for superresolution. The same argument can be applied to determining the separation of two closely spaced emitters: the total Fisher information carried by N photons on the separation is \underline{F}=NF_{\text{sep,1phot}} , where F_{\text{sep,1phot}} is the information on the separation carried by one photon. This information tends to zero together with the separation of the emitters [ 22 ] , however, for whatever small F_{\text{sep,1phot}} , one can make the total information arbitrarily large by collecting sufficiently large number of photons and thus determine the separation with arbitrary precision. In practice, a fluorophore emits a limited number of photons before photobleaching, and, therefore, individual localization of blinking emitters gives much better results than a simultaneous localization of multiple cofluorescent emitters within the width of PSF. The precision gain attained by making the sources blink is the key feature of SMLM. A question arises: Is it a general law of Nature that blinking emitters send more information on their locations than cofluorescent ones? In this paper, we give a positive answer to this question, showing that the informational advantage gained by blinking is a consequence of another fundamental property of Fisher information, its convexity . We consider the task of localization of an arbitrary number of light sources as a multiparameter estimation problem and prove that, in a scenario where all sources emit the same number of photons and microscopy satisfies the conditions of translation and rotation invariance, the Fisher information matrix possesses the property of weak matrix convexity , which results in a higher localization efficiency for blinking sources. We also prove, that in the general case of arbitrary numbers of emitted photons, the Fisher information matrix possesses the property of trace convexity , which leads to a higher localization efficiency of blinking sources when it is measured in terms of the new measure, which we call “average eigenparameter precision.” The paper is structured as follows. In Sec. II we recast the formalism of quantum metrology [ 23 , 24 , 25 , 26 ] in the context

Read the complete text