Temperature dependence in Krylov space

Nikolaos Angelinos, Debarghya Chakraborty, Anatoly Dymarsky · 2025

Read Temperature dependence in Krylov space by Nikolaos Angelinos, Debarghya Chakraborty, Anatoly Dymarsky (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-2508-19233.txt.

Introduction

Nikolaos Angelinos, Debarghya Chakraborty, and Anatoly Dymarsky wrote this 2025 physics paper on how temperature enters a method for tracking quantum dynamics. The recursion method builds a chain of coefficients, called Lanczos coefficients, from a system's two-point function. How those numbers shift with temperature was not understood in general.

The authors show that the temperature dependence is governed by integrable dynamics, a class of motion with exact structure. After a change of variables, the even and odd coefficients follow two linked Toda chains. Checking that the two chains agree yields a test they call the Krylov bootstrap. That test forces some scale-invariant models to have a degenerate spectrum. Known late-time behavior of the chains then gives control over the system at very low temperature.

Catalog record

CREATORNikolaos Angelinos, Debarghya Chakraborty, Anatoly Dymarsky
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE81 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-19233

Browse the full library · About the Tower of Babel

Full text

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

Temperature dependence in Krylov space Nikolaos Angelinos, Debarghya Chakraborty, Anatoly Dymarsky Abstract We consider the recursion method applied to a generic 2pt function of a quantum system and show, in full generality, that the temperature dependence of the corresponding Lanczos coefficients is governed by integrable dynamics. After an appropriate change of variables, Lanczos coefficients with even and odd indices are described by two independent Toda chains, related at the level of the initial conditions. Consistency of the resulting equations can be used to show that certain scale-invariant models necessarily have a degenerate spectrum. We dub this self-consistency-based approach the “Krylov bootstrap”. The known analytic behavior of the Toda chain at late times translates into analytic control over the 2pt function and Krylov complexity at very low temperatures. We also discuss the behavior of Lanczos coefficients when the temperature is low but not much smaller than the spectral gap, and elucidate the origin of the staggering behavior of Lanczos coefficients in this regime. 1 Introduction The Krylov space method, and Krylov complexity in particular, have emerged recently as a new way to explore and characterize quantum chaotic dynamics NANDY20251 ; Rabinovici:2025otw . The Krylov space is defined as the minimal vector space that contains an operator A , together with its time evolution A(t) . It is spanned by nested commutators of the operator with the Hamiltonian [H,[H,\cdots[H,A]]] . Utilizing the recursion method, the two-point function of A , that we will denote C(t) , can be rewritten in terms of the so-called Lanczos coefficients \{b_{n}\} , that fully encode and characterize it. Although the content of C(t) and b_{n} are mathematically equivalent, the behavior of b_{n} was shown to offer various insights into system’s dynamics Parker19 . Thus, Krylov complexity K(t) , which is essentially the mean location of the operator A(t) as it spreads along the “Krylov chain” – a certain preferred basis in Krylov space – has been proposed to characterize complexity growth and is conjectured to bound the OTOC growth Parker19 , generalizing the Maldacena-Shenker-Stanford bound on chaos avdoshkin2022krylov . Considering a system at finite temperature T=\beta^{-1} , Lanczos coefficients will exhibit temperature dependence b_{n}(\beta) that encodes temperature dependence of the autocorrelation function C_{\beta}(t) . As one of the central results of this paper we find that \beta dependence of b_{n} is described by a completely integrable dynamical system, namely two uncoupled Toda chains, related at the level of initial conditions. This result is conceptually and technically similar to todaKrylov , where the dependence of Lanczos coefficients on Euclidean time was found to be governed by the Toda lattice equations. An explicit form of the differential equations governing \beta -dependence of b_{n} implies various consistency conditions, giving rise to “Krylov bootstrap”. These differential equations can also be used for numerical integration, calculating the thermal two-point function C_{\beta}(t) starting from the data associated with any other inverse temperature \beta_{0} . Furthermore, recasting \beta -dynamics in terms of a completely integrable dynamical system provides analytic control in the large- \beta limit, \beta m\gg 1 , where m is the spectral gap of the system. In this regime we find the asymptotic value of Krylov complexity at late times to behave as K\sim e^{-\beta m/2} . We also discuss the regime where \beta is large but not much larger than m^{-1} . The integrable hierarchy does not yield fully universal predictions here, but we employ a number of exact mathematical results to explain the origin of the so-called staggering behavior, where b_{n} splits into two smooth branches, observed empirically earlier in Mitra ; Mitra2 ; Mitra3 ; Yeh:2023fek ; avdoshkin2022krylov ; Camargo_2023 . The paper is organized as follows. In 2 we derive the analytic equations driving the \beta -dependence of b_{n} . Section 3 makes use of these results by studying the large temperature limit, as well as evaluating C_{\beta}(t) numerically starting from \beta=0 data. Section 4 is devoted to the moderately large- \beta regime, where we use simple analytic models to describe characteristic behavior of b_{n} exhibited by many spatially-extended systems. We conclude with a discussion in section 5 . 2 Temperature dependence of Lanczos coefficients as a completely integrable system The starting point is the Lanczos algorithm, which is used to iteratively build an orthogonal basis in the Krylov space, starting with an initial operator A . Orthogonality is defined with respect to a choice of an inner product, defined in terms of the correlation function. Our goal is to understand temperature dependence of the correlation function and we define the scalar product \langle A,B\rangle_{\beta}\equiv\Tr(A^\dagger\rho_1(\beta) B\rho_2(\beta) ). (1) such that the autocorrelation function C_{\beta}(t) and spectral function \Phi(\omega) are given by C_{\beta}(t)=\langle A(0),A(t)\rangle_{\beta}=\int_{-\infty}^{\infty}d\omega~\Phi(\omega)e^{i\omega t}. (2) In general \rho_{1}(\beta) and \rho_{2}(\beta) are Hermitian positive semi-definite operators that commute with the system’s Hamiltonian. In most of this work we focus on the Wightman inner product by making the choice \rho_{1}=\rho_{2}={e^{-{\beta\over 2}H}}, (3) where H is the Hamiltonian of the quantum system. Later in 2.3 , we will explain how our formalism can be straightforwardly adapted to the choice of other thermal inner products. For an initial Hermitian operator A_{0}=A , we define the Krylov basis recursively A_{n+1}=[H,A_{n}]-b_{n-1}^{2}A_{n-1}, (4) where we set b_{-1}=0 . The Lanczos coefficients b_{n}(\beta) are fixed by requiring that the basis is orthogonal, b_{n}^{2}(\beta)={\langle A_{n+1},A_{n+1}\rangle_{\beta}\over\langle A_{n},A_{n}\rangle_{\beta}}. (5) Since the norms are non-negative, for convenience, we define q_{n}(\beta) as follows \delta_{nm}e^{q_{n}(\beta)}:=\langle A_{n},A_{m}\rangle_{\beta}. (6) It is sometimes convenient to work with the orthonormal Krylov basis \{O_{n}\} defined by O_{n}(\beta):={A_{n}\over\sqrt{\langle A_{n},A_{n}\rangle}_{\beta}}={A_{n}}(\beta)e^{-{q_{n}(\beta)\over 2}}. (7) Since the inner product ( 3 ) is \beta -dependent, all quantities introduced above, including the Lanczos coefficients b_{n}(\beta) and the Krylov basis operators O_{n}(\beta) , are \beta -dependent. Our main goal is to understand this temperature dependence. We will often suppress the explicit \beta -dependence in our notation, except where necessary for clarity. From ( 4 ) it follows that the action of the Liouvillian [H,\cdot] on the normalized Krylov basis has a tridiagonal representation, M_{nm}(\beta):=\langle O_{n}(\beta),[H,O_{m}(\beta)]\rangle_{\beta}=\begin{pmatrix}0&b_{0}&0&0&\ddots\\ b_{0}&0&b_{1}&0&\ddots\\ 0&b_{1}&0&b_{2}&\ddots\\ 0&0&b_{2}&0&\ddots\\ \ddots&\ddots&\ddots&\ddots&\ddots\end{pmatrix}. (8) Central to our work is the point that regardless of the temperature, which defines the scalar product in the Krylov space, the Krylov space itself, and the adjoint action of H are temperature-independent. This implies then that M(\beta) for different values of \beta are different representations of the same operator, thus \beta -dependence is an isospectral deformation of M which we will describe in terms of integrable dynamics. To this end it is convenient to introduce “temperature evolution” by considering \beta -dependent operators, D[\beta]:=e^{-\frac{\beta}{2}H}\,D\,e^{-\frac{\beta}{2}H}, (9) for any operator D . An important subtlety is that temperature evolution of A=A_{0} under ( 9 ) will in most cases move it outside of the Krylov space. To see that, we decompose A as follows, A=A^{\perp}+A_{\rm d}, (10) where A_{\rm d} and A^{\perp} stand for diagonal and off-diagonal parts of A , written in the eigenbasis of H . The diagonal operator A_{\rm d} is in the nullspace of the Liouvillian and therefore all operators A_{n} defined by ( 4 ) with the same parity of n will share the same diagonal part (up to normalization). On the contrary, the diagonal part of A[\beta] is \beta -dependent and hence for \beta\neq 0 is outside the Krylov

Read the complete text