# Bryan’s Maximum Entropy Method—Diagnosis of a Flawed Argument and Its Remedy

## Abstract

**:**

## 1. Introduction

## 2. Diagnosis of the Problem in Bryan’s MEM

#### 2.1. Tikhonov Regularization

#### 2.2. Maximum Entropy Method

#### 2.3. Numerical Evidence for the Inadequacy of the SVD Subspace

## 3. Remedy of the Problem

#### 3.1. Staying within MEM

#### 3.2. Beyond MEM

## 4. Summary and Conclusions

## Author Contributions

## Funding

## Conflicts of Interest

## Abbreviations

BFGS | Broyden–Fletcher–Goldfarb–Shanno |

BR | Bayesian Reconstruction |

MEM | Maximum Entropy Method |

SJ | Shannon–Jaynes |

SVD | singular value decomposition |

TK | Tikhonov |

## Appendix A. Explicit SVD Example of (Non-) Projection to the Null-Space

## Appendix B. Explicit MEM Example with a Solution Outside of the SVD Subspace

## References

- Whittaker, E.T. On a New Method of Graduation. Proc. Edinb. Math. Soc.
**1922**, 41, 63–75. [Google Scholar] [CrossRef] [Green Version] - Tikhonov, A.N. On the stability of inverse problems. Dokl. Akad. Nauk SSSR
**1943**, 39, 195–198. [Google Scholar] - Narayan, R.; Nityananda, R. Maximum Entropy Image Restoration in Astronomy. Annu. Rev. Astron. Astrophys.
**1986**, 24, 127–170. [Google Scholar] [CrossRef] - Jarrell, M.; Gubernatis, J.E. Bayesian inference and the analytic continuation of imaginary-time quantum Monte Carlo data. Phys. Rep.
**1996**, 269, 133–195. [Google Scholar] [CrossRef] [Green Version] - Burnier, Y.; Rothkopf, A. Bayesian Approach to Spectral Function Reconstruction for Euclidean Quantum Field Theories. Phys. Rev. Lett.
**2013**, 111, 182003. [Google Scholar] [CrossRef] [Green Version] - Asakawa, M.; Nakahara, Y.; Hatsuda, T. Maximum entropy analysis of the spectral functions in lattice QCD. Prog. Part. Nucl. Phys.
**2001**, 46, 459–508. [Google Scholar] [CrossRef] [Green Version] - Bryan, R.K. Maximum entropy analysis of oversampled data problems. Eur. Biophys. J.
**1990**, 18, 165–174. [Google Scholar] [CrossRef] - Asakawa, M. Comment on “Heavy Quarkonium in Extreme Conditions”. arXiv
**2020**, arXiv:2001.10205. [Google Scholar] - Rothkopf, A. Heavy quarkonium in extreme conditions. Phys. Rep.
**2020**. [Google Scholar] [CrossRef] - Rothkopf, A. Improved maximum entropy analysis with an extended search space. J. Comput. Phys.
**2013**, 238, 106–114. [Google Scholar] [CrossRef] [Green Version] - Jakovác, A.; Petreczky, P.; Petrov, K.; Velytsky, A. Quarkonium correlators and spectral functions at zero and finite temperature. Phys. Rev. D
**2007**, 75, 014506. [Google Scholar] [CrossRef] [Green Version] - Rothkopf, A.; Hatsuda, T.; Sasaki, S. Complex Heavy-Quark Potential at Finite Temperature from Lattice QCD. Phys. Rev. Lett.
**2012**, 108, 162001. [Google Scholar] [CrossRef] [Green Version] - Rothkopf, A. Improved Maximum Entropy Method with an Extended Search Space. In Proceedings of the 30th International Symposium on Lattice Field Theory (LATTICE 2012), Cairns, Australia, 24–29 June 2012; p. 100. [Google Scholar] [CrossRef] [Green Version]
- Kelly, A.; Rothkopf, A.; Skullerud, J.I. Bayesian study of relativistic open and hidden charm in anisotropic lattice QCD. Phys. Rev. D
**2018**, 97, 114509. [Google Scholar] [CrossRef] [Green Version] - Sandvik, A.W. Stochastic method for analytic continuation of quantum Monte Carlo data. Phys. Rev. B
**1998**, 57, 10287–10290. [Google Scholar] [CrossRef] - Beach, K.S.D. Identifying the maximum entropy method as a special limit of stochastic analytic continuation. arXiv
**2004**, arXiv:Cond-mat/0403055. [Google Scholar] - Mishchenko, A.S.; Prokof’ev, N.V.; Sakamoto, A.; Svistunov, B.V. Diagrammatic quantum Monte Carlo study of the Fr\ohlich polaron. Phys. Rev. B
**2000**, 62, 6317–6336. [Google Scholar] [CrossRef] - Ding, H.T.; Kaczmarek, O.; Mukherjee, S.; Ohno, H.; Shu, H.T. Stochastic reconstructions of spectral functions: Application to lattice QCD. Phys. Rev. D
**2018**, 97, 094503. [Google Scholar] [CrossRef] [Green Version] - Yoon, H.; Sim, J.H.; Han, M.J. Analytic continuation via domain knowledge free machine learning. Phys. Rev. B
**2018**, 98, 245101. [Google Scholar] [CrossRef] [Green Version] - Fournier, R.; Wang, L.; Yazyev, O.V.; Wu, Q. Artificial Neural Network Approach to the Analytic Continuation Problem. Phys. Rev. Lett.
**2020**, 124, 056401. [Google Scholar] [CrossRef] [Green Version] - Kades, L.; Pawlowski, J.M.; Rothkopf, A.; Scherzer, M.; Urban, J.M.; Wetzel, S.J.; Wink, N.; Ziegler, F. Spectral Reconstruction with Deep Neural Networks. arXiv
**2019**, arXiv:1905.04305. [Google Scholar] - Offler, S.; Aarts, G.; Allton, C.; Glesaaen, J.; Jäger, B.; Kim, S.; Lombardo, M.P.; Ryan, S.M.; Skullerud, J.I. News from bottomonium spectral functions in thermal QCD. arXiv
**2019**, arXiv:1912.12900. [Google Scholar] - Rothkopf, A. Bayesian techniques and applications to QCD. In Proceedings of the 13th Conference on Quark Confinement and the Hadron Spectrum (Confinement XIII), Maynooth, Ireland, 31 July–6 August 2018; p. 26. [Google Scholar]

**Figure 1.**The mock function $\rho $ (

**left**) and corresponding mock data (

**right**) deployed in the explicit example discussed in the main text, which shows that the extremum of the posterior is not necessarily located within the singular value decomposition (SVD) subspace.

**Figure 2.**The SVD basis functions ${U}_{1}\left(\omega \right)$, ${U}_{2}\left(\omega \right)$, and ${U}_{3}\left(\omega \right)$, which Bryan’s argument suggests should capture the extremum of the posterior for all three mock functions ${\rho}_{i}$ after exponentiation. Note that the SVD basis functions flatten out about $\omega =10$. Their domain extends to ${\omega}_{\mathrm{max}}=1000$ but their value stays close to zero.

**Figure 3.**Comparison of the behavior of the integrand of the Tikhonov (TK), MEM, and Bayesian Reconstruction (BR) regulators for the choice $\alpha =0.1$ and $m=1$ using (

**left**) a log-lin scale and (

**right**) a log-log scale. Note that the flattening off of the Shannon–Jaynes entropy towards vanishing $\rho $.

© 2020 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Rothkopf, A.
Bryan’s Maximum Entropy Method—Diagnosis of a Flawed Argument and Its Remedy. *Data* **2020**, *5*, 85.
https://doi.org/10.3390/data5030085

**AMA Style**

Rothkopf A.
Bryan’s Maximum Entropy Method—Diagnosis of a Flawed Argument and Its Remedy. *Data*. 2020; 5(3):85.
https://doi.org/10.3390/data5030085

**Chicago/Turabian Style**

Rothkopf, Alexander.
2020. "Bryan’s Maximum Entropy Method—Diagnosis of a Flawed Argument and Its Remedy" *Data* 5, no. 3: 85.
https://doi.org/10.3390/data5030085