Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2607.17346 · Physical review. D/Physical review. D.

The Information Content of Krylov Observables: A Machine Learning Approach

Ritam Basu

hep-thcond-mat.dis-nncond-mat.stat-mechmath-phquant-phSubmitted Jul 19, 2026• Announced Jul 21, 2026• 0 citations• Revised Oct 12, 2026 (v2)
v1 · Jul 19, 2026v2 · Oct 12, 2026
arXivHTMLPDF

Abstract

We employ machine learning techniques to estimate the amount of information contained in three Krylov-space observables: the spread complexity $C(t)$, the discrete Wigner negativity $N(t)$, and the normalized negativity $χ(t)=N(t)/|S(t)|$, where $S(t)$ is the survival amplitude. We study thermofield-double evolutions on GUE, GOE and Poisson spectra, and on the integrable $SL(2,R)$ Krylov chain of a two-dimensional CFT primary, with Lanczos coefficients $b_n=α\sqrt{n(n+2h-1)}$. To introduce chaos we add a GUE random matrix to the integrable chain, $H(\varepsilon)=H_{SL(2,R)}+\varepsilon R_0 W_{GUE}$, where $R_0$ is its spectral half-width. Either $C(t)$ or $N(t)$ alone determines the thermofield temperature, with $R^2\simeq0.999$. However, neither can reconstruct the fine structure of the spectral form factor. In the integrable sector all three observables carry the same information. When we turn on chaos an asymmetry appears: $χ$ predicts $C$ better than $C$ predicts $χ$. This gap grows from $+0.33$ to $+0.77$, while the same gap for the raw negativity decays to zero. The extra information comes from the survival amplitude, and $χ(t)$ carries it while $C(t)$ can still be recovered from $χ(t)$. We derive a lower bound on this gap, set by the fraction of the variance of $χ$ that comes from $S$.

DOI
10.1103/hqts-lp2n
Comments
54 pages, 34 figures, 4 tables. Accepted for publication in Physical Review D, https://journals.aps.org/prd/accepted/10.1103/hqts-lp2n

Related papers

The Normalised Wigner Negativity Rate as a Second-Moment Probe of Infall in AdS$_3$

Ritam Basu

In spread complexity, the growth rate of the average position of a time-evolved state along its Krylov chain recovers the proper radial momentum of an infalling particle in AdS. It is, however, a measure of the first moment only, irrespective of the spread of the wavepacket away from its classical trajectory. The rate of a normalized Krylov-Wigner negativity can be proposed as a diagnostic of the second moment of the boundary state that captures this spreading. Starting with the seed-normalized Krylov-Wigner distribution, that is, the Wigner transform of the descendant cloud with the decaying return amplitude divided out, we obtain an analytic Bessel form in the macroscopic limit and compute its total negativity explicitly. Retaining the Bessel variable all the way through, we find that the negativity goes as $\sinh^{4Δ}(πt/β)$. But the raw normalized-state negativity saturates to an $O(1)$ constant, well below the $O(\sqrt{D})$ bound. Using the exact negative binomial statistics of the Krylov chain, the normalized negativity is at late times a fixed power of the second moment of the Krylov wavepacket, $\mathcal{N}(t)\propto[\mathrm{Var}(\hat{N}_{\mathrm{Krylov}})]^Δ$, for every $Δ>1/2$. The relation linearizes precisely at $Δ=1$. In this regime via the momentum dictionary of Caputa et al. (arXiv:2410.23334), the rate becomes the product of the proper radial position and momentum, $\mathcal{R}\propto\mathcal{C}\,P_ρ$, which we interpret as the rate of the tidal stretch of nearby geodesics falling into the horizon. We also comment briefly on a speculative direction for future research, in particular the interpretation of the transverse string size operator in terms of the Krylov number operator through the common $\mathrm{SU}(1,1)$ discrete series.

hep-thCross-listedarXiv:2607.04065•Physical review. D/Physical review. D.•Jul 7, 2026•3 citations

Krylov complexity of density matrix operators

Paweł Caputa, Hyun-Sik Jeong, Sinong Liu +2 more

A bstract Quantifying complexity in quantum systems has witnessed a surge of interest in recent years, with Krylov-based measures such as Krylov complexity ( C K ) and Spread complexity ( C S ) gaining prominence. In this study, we investigate their interplay by considering the complexity of states represented by density matrix operators . After setting up the problem, we analyze a handful of analytical and numerical examples spanning generic two-dimensional Hilbert spaces, qubit states, quantum harmonic oscillators, and random matrix theories, uncovering insightful relationships. For generic pure states, our analysis reveals two key findings: (I) a correspondence between moment-generating functions (of Lanczos coefficients) and survival amplitudes, and (II) an early-time equivalence between C K and 2 C S . Furthermore, for maximally entangled pure states, we find that the moment-generating function of C K becomes the Spectral Form Factor and, at late-times, C K is simply related to NC S for N ≥ 2 within the N -dimensional Hilbert space. Notably, we confirm that C K = 2 C S holds across all times when N = 2. Through the lens of random matrix theories, we also discuss deviations between complexities at intermediate times and highlight subtleties in the averaging approach at the level of the survival amplitude.

arXiv:2402.09522•Journal of High Energy Physics•May 31, 2024•91 citations

Operator Krylov complexity in random matrix theory

Tang Haifeng

Krylov complexity, as a novel measure of operator complexity under Heisenberg evolution, exhibits many interesting universal behaviors and also bounds many other complexity measures. In this work, we study Krylov complexity $\mathcal{K}(t)$ in Random Matrix Theory (RMT). In large $N$ limit: (1) For infinite temperature, we analytically show that the Lanczos coefficient $\{b_n\}$ saturate to constant plateau $\lim\limits_{n\rightarrow\infty}b_n=b$, rendering a linear growing complexity $\mathcal{K}(t)\sim t$, in contrast to the exponential-in-time growth in chaotic local systems in thermodynamic limit. After numerically comparing this plateau value $b$ to a large class of chaotic local quantum systems, we find that up to small fluctuations, it actually bounds the $\{b_n\}$ in chaotic local quantum systems. Therefore we conjecture that in chaotic local quantum systems after scrambling time, the speed of linear growth of Krylov complexity cannot be larger than that in RMT. (2) For low temperature, we analytically show that $b_n$ will first exhibit linear growth with $n$, whose slope saturates the famous chaos bound. After hitting the same plateau $b$, $b_n$ will then remain constant. This indicates $\mathcal{K}(t)\sim e^{2πt/β}$ before scrambling time $t_*\sim O(β\logβ)$, and after that it will grow linearly in time, with the same speed as in infinite temperature. We finally remark on the effect of finite $N$ corrections.

arXiv:2312.17416•Dec 29, 2023•19 citations

Operator K-complexity in DSSYK: Krylov complexity equals bulk length

Marco Ambrosini, Eliezer Rabinovici, Adrián Sánchez-Garrido +2 more

A bstract In this paper we study the notion of complexity under time evolution in chaotic quantum systems with holographic duals. Continuing on from our previous work, we turn our attention to the issue of Krylov complexity upon the insertion of a class of single-particle operators in the double-scaled SYK model. Such an operator is described by a matter-chord insertion, which splits the theory into left/right sectors, allowing us, via chord-diagram technology, to compute two different notions of complexity associated to the operator insertion: first a Krylov operator complexity, and second the Krylov complexity of a state obtained by an operator acting on the thermofield double state. We will provide both an analytic proof and detailed numerical evidence, that both Krylov complexities arise from a recursively defined basis of states characterized by a constant total chord number. As a consequence, in all cases we are able to establish that Krylov complexity is given by the expectation value of a length operator acting on the Hilbert space of the theory, expressed in terms of basis states, organized by left and right chord number. We find analytic expressions for the semiclassical limit of K-complexity, and study how the size of the operator encodes the scrambling dynamics upon the matter insertion in Krylov language. We furthermore determine the effective Hamiltonian governing the evolution of K-complexity, showing that evolution on the Krylov chain can equivalently be understood as a particle moving in a Morse potential. A particular type of triple scaling limit allows to access the gravitational sector of the theory, in which the geometrical nature of K-complexity is assured by virtue of being a total chord length, in an analogous fashion to what was found in [1] for the K-complexity of the thermofield double state.

arXiv:2412.15318•Journal of High Energy Physics•Aug 7, 2025•55 citations

Krylov complexity in mixed phase space

Kyoung-Bum Huh, Hyun-Sik Jeong, Leopoldo A. Pando Zayas +1 more

We investigate the Krylov complexity of thermofield double states in systems with mixed phase space, uncovering a direct correlation with the Brody distribution, which interpolates between Poisson and Wigner statistics. Our analysis spans two-dimensional random matrix models featuring (I) GOE-Poisson and (II) GUE-Poisson transitions and extends to higher-dimensional cases, including a stringy matrix model (GOE-Poisson) and the mass-deformed SYK model (GUE-Poisson). Krylov complexity consistently emerges as a reliable marker of quantum chaos, displaying a characteristic peak in the chaotic regime that gradually diminishes as the Brody parameter approaches zero, signaling a shift toward integrability. These results establish Krylov complexity as a powerful diagnostic of quantum chaos and highlight its interplay with eigenvalue statistics in mixed phase systems.

arXiv:2412.04963•Physical review. D/Physical review. D.•May 21, 2025•31 citations

Onset of random-matrix statistics in the Yukawa-Sachdev-Ye-Kitaev model: Spectral correlations and Krylov complexity

Sizheng Cao, Hyun-Sik Jeong, Yi-Li Wang

The Yukawa-Sachdev-Ye-Kitaev (YSYK) model, a solvable model of non-Fermi liquids with a proposed cavity-QED realization, couples $N$ complex fermions to $M$ bosons of frequency $ω_0$ through random Yukawa couplings of strength $g$, and its dynamics depends only on $R\equivω_0/g^{2/3}$. For $R\ll1$ the bosons are slow and act as a static random background in which the fermions are free, while for $R\gg1$ they are only virtually excited and generate a random four-fermion interaction of rank $M$, which is generically chaotic for $M\geq2$. Using exact diagonalization at infinite temperature, we follow the emergence of random-matrix statistics between these limits with unfolded spectral statistics and the Krylov complexity of the thermofield-double state. For complex couplings the level-spacing distribution evolves from Poisson to Gaussian-unitary-ensemble (GUE) statistics, the spectral form factor develops a correlation hole and a ramp, and the Krylov complexity evaluated on the unfolded spectrum develops a peak above its late-time plateau whose height reaches the GUE value. All diagnostics locate the onset around $R\approx0.3$, and adding boson modes moves it to smaller $R$. This delineates the parameter regime in which a cavity-QED realization would display random-matrix spectral correlations. On the other hand, real couplings give the same crossover towards Gaussian-orthogonal-ensemble (GOE) statistics, and along an interpolation between the two the peak height follows the mean gap ratio, so the peak identifies the symmetry class as well as the onset of chaos. Without unfolding the peak stays about 20$\%$ below its random-matrix value in both classes, which we trace to the non-semicircular density of states, and at larger $R$ the boson-number bands slow down the spreading of the state in Krylov space.

nlin.CDCross-listedarXiv:2609.30333•Sep 25, 2026•0 citations
We employ machine learning techniques to estimate the amount of information contained in three Krylov-space observables: the spread complexity C(t)C(t)C(t), the discrete Wigner negativity N(t)N(t)N(t), and the normalized negativity χ(t)=N(t)/∣S(t)∣χ(t)=N(t)/|S(t)|χ(t)=N(t)/∣S(t)∣, where S(t)S(t)S(t) is the survival amplitude. We study thermofield-double evolutions on GUE, GOE and Poisson spectra, and on the integrable SL(2,R)SL(2,R)SL(2,R) Krylov chain of a two-dimensional CFT primary, with Lanczos coefficients bn=αn(n+2h−1)b_n=α\sqrt{n(n+2h-1)}bn​=αn(n+2h−1)​. To introduce chaos we add a GUE random matrix to the integrable chain, H(ε)=HSL(2,R)+εR0WGUEH(\varepsilon)=H_{SL(2,R)}+\varepsilon R_0 W_{GUE}H(ε)=HSL(2,R)​+εR0​WGUE​, where R0R_0R0​ is its spectral half-width. Either C(t)C(t)C(t) or N(t)N(t)N(t) alone determines the thermofield temperature, with R2≃0.999R^2\simeq0.999R2≃0.999. However, neither can reconstruct the fine structure of the spectral form factor. In the integrable sector all three observables carry the same information. When we turn on chaos an asymmetry appears: χχχ predicts CCC better than CCC predicts χχχ. This gap grows from +0.33+0.33+0.33 to +0.77+0.77+0.77, while the same gap for the raw negativity decays to zero. The extra information comes from the survival amplitude, and χ(t)χ(t)χ(t) carries it while C(t)C(t)C(t) can still be recovered from χ(t)χ(t)χ(t). We derive a lower bound on this gap, set by the fraction of the variance of χχχ that comes from SSS.
The Normalised Wigner Negativity Rate as a Second-Moment Probe of Infall in AdS3_33​
In spread complexity, the growth rate of the average position of a time-evolved state along its Krylov chain recovers the proper radial momentum of an infalling particle in AdS. It is, however, a measure of the first moment only, irrespective of the spread of the wavepacket away from its classical trajectory. The rate of a normalized Krylov-Wigner negativity can be proposed as a diagnostic of the second moment of the boundary state that captures this spreading. Starting with the seed-normalized Krylov-Wigner distribution, that is, the Wigner transform of the descendant cloud with the decaying return amplitude divided out, we obtain an analytic Bessel form in the macroscopic limit and compute its total negativity explicitly. Retaining the Bessel variable all the way through, we find that the negativity goes as sinh⁡4Δ(πt/β)\sinh^{4Δ}(πt/β)sinh4Δ(πt/β). But the raw normalized-state negativity saturates to an O(1)O(1)O(1) constant, well below the O(D)O(\sqrt{D})O(D​) bound. Using the exact negative binomial statistics of the Krylov chain, the normalized negativity is at late times a fixed power of the second moment of the Krylov wavepacket, N(t)∝[Var(N^Krylov)]Δ\mathcal{N}(t)\propto[\mathrm{Var}(\hat{N}_{\mathrm{Krylov}})]^ΔN(t)∝[Var(N^Krylov​)]Δ, for every Δ>1/2Δ>1/2Δ>1/2. The relation linearizes precisely at Δ=1Δ=1Δ=1. In this regime via the momentum dictionary of Caputa et al. (arXiv:2410.23334), the rate becomes the product of the proper radial position and momentum, R∝C Pρ\mathcal{R}\propto\mathcal{C}\,P_ρR∝CPρ​, which we interpret as the rate of the tidal stretch of nearby geodesics falling into the horizon. We also comment briefly on a speculative direction for future research, in particular the interpretation of the transverse string size operator in terms of the Krylov number operator through the common SU(1,1)\mathrm{SU}(1,1)SU(1,1) discrete series.
A bstract Quantifying complexity in quantum systems has witnessed a surge of interest in recent years, with Krylov-based measures such as Krylov complexity ( C K ) and Spread complexity ( C S ) gaining prominence. In this study, we investigate their interplay by considering the complexity of states represented by density matrix operators . After setting up the problem, we analyze a handful of analytical and numerical examples spanning generic two-dimensional Hilbert spaces, qubit states, quantum harmonic oscillators, and random matrix theories, uncovering insightful relationships. For generic pure states, our analysis reveals two key findings: (I) a correspondence between moment-generating functions (of Lanczos coefficients) and survival amplitudes, and (II) an early-time equivalence between C K and 2 C S . Furthermore, for maximally entangled pure states, we find that the moment-generating function of C K becomes the Spectral Form Factor and, at late-times, C K is simply related to NC S for N ≥ 2 within the N -dimensional Hilbert space. Notably, we confirm that C K = 2 C S holds across all times when N = 2. Through the lens of random matrix theories, we also discuss deviations between complexities at intermediate times and highlight subtleties in the averaging approach at the level of the survival amplitude.
Krylov complexity, as a novel measure of operator complexity under Heisenberg evolution, exhibits many interesting universal behaviors and also bounds many other complexity measures. In this work, we study Krylov complexity K(t)\mathcal{K}(t)K(t) in Random Matrix Theory (RMT). In large NNN limit: (1) For infinite temperature, we analytically show that the Lanczos coefficient {bn}\{b_n\}{bn​} saturate to constant plateau lim⁡n→∞bn=b\lim\limits_{n\rightarrow\infty}b_n=bn→∞lim​bn​=b, rendering a linear growing complexity K(t)∼t\mathcal{K}(t)\sim tK(t)∼t, in contrast to the exponential-in-time growth in chaotic local systems in thermodynamic limit. After numerically comparing this plateau value bbb to a large class of chaotic local quantum systems, we find that up to small fluctuations, it actually bounds the {bn}\{b_n\}{bn​} in chaotic local quantum systems. Therefore we conjecture that in chaotic local quantum systems after scrambling time, the speed of linear growth of Krylov complexity cannot be larger than that in RMT. (2) For low temperature, we analytically show that bnb_nbn​ will first exhibit linear growth with nnn, whose slope saturates the famous chaos bound. After hitting the same plateau bbb, bnb_nbn​ will then remain constant. This indicates K(t)∼e2πt/β\mathcal{K}(t)\sim e^{2πt/β}K(t)∼e2πt/β before scrambling time t∗∼O(βlog⁡β)t_*\sim O(β\logβ)t∗​∼O(βlogβ), and after that it will grow linearly in time, with the same speed as in infinite temperature. We finally remark on the effect of finite NNN corrections.
A bstract In this paper we study the notion of complexity under time evolution in chaotic quantum systems with holographic duals. Continuing on from our previous work, we turn our attention to the issue of Krylov complexity upon the insertion of a class of single-particle operators in the double-scaled SYK model. Such an operator is described by a matter-chord insertion, which splits the theory into left/right sectors, allowing us, via chord-diagram technology, to compute two different notions of complexity associated to the operator insertion: first a Krylov operator complexity, and second the Krylov complexity of a state obtained by an operator acting on the thermofield double state. We will provide both an analytic proof and detailed numerical evidence, that both Krylov complexities arise from a recursively defined basis of states characterized by a constant total chord number. As a consequence, in all cases we are able to establish that Krylov complexity is given by the expectation value of a length operator acting on the Hilbert space of the theory, expressed in terms of basis states, organized by left and right chord number. We find analytic expressions for the semiclassical limit of K-complexity, and study how the size of the operator encodes the scrambling dynamics upon the matter insertion in Krylov language. We furthermore determine the effective Hamiltonian governing the evolution of K-complexity, showing that evolution on the Krylov chain can equivalently be understood as a particle moving in a Morse potential. A particular type of triple scaling limit allows to access the gravitational sector of the theory, in which the geometrical nature of K-complexity is assured by virtue of being a total chord length, in an analogous fashion to what was found in [1] for the K-complexity of the thermofield double state.
We investigate the Krylov complexity of thermofield double states in systems with mixed phase space, uncovering a direct correlation with the Brody distribution, which interpolates between Poisson and Wigner statistics. Our analysis spans two-dimensional random matrix models featuring (I) GOE-Poisson and (II) GUE-Poisson transitions and extends to higher-dimensional cases, including a stringy matrix model (GOE-Poisson) and the mass-deformed SYK model (GUE-Poisson). Krylov complexity consistently emerges as a reliable marker of quantum chaos, displaying a characteristic peak in the chaotic regime that gradually diminishes as the Brody parameter approaches zero, signaling a shift toward integrability. These results establish Krylov complexity as a powerful diagnostic of quantum chaos and highlight its interplay with eigenvalue statistics in mixed phase systems.
The Yukawa-Sachdev-Ye-Kitaev (YSYK) model, a solvable model of non-Fermi liquids with a proposed cavity-QED realization, couples NNN complex fermions to MMM bosons of frequency ω0ω_0ω0​ through random Yukawa couplings of strength ggg, and its dynamics depends only on R≡ω0/g2/3R\equivω_0/g^{2/3}R≡ω0​/g2/3. For R≪1R\ll1R≪1 the bosons are slow and act as a static random background in which the fermions are free, while for R≫1R\gg1R≫1 they are only virtually excited and generate a random four-fermion interaction of rank MMM, which is generically chaotic for M≥2M\geq2M≥2. Using exact diagonalization at infinite temperature, we follow the emergence of random-matrix statistics between these limits with unfolded spectral statistics and the Krylov complexity of the thermofield-double state. For complex couplings the level-spacing distribution evolves from Poisson to Gaussian-unitary-ensemble (GUE) statistics, the spectral form factor develops a correlation hole and a ramp, and the Krylov complexity evaluated on the unfolded spectrum develops a peak above its late-time plateau whose height reaches the GUE value. All diagnostics locate the onset around R≈0.3R\approx0.3R≈0.3, and adding boson modes moves it to smaller RRR. This delineates the parameter regime in which a cavity-QED realization would display random-matrix spectral correlations. On the other hand, real couplings give the same crossover towards Gaussian-orthogonal-ensemble (GOE) statistics, and along an interpolation between the two the peak height follows the mean gap ratio, so the peak identifies the symmetry class as well as the onset of chaos. Without unfolding the peak stays about 20%\%% below its random-matrix value in both classes, which we trace to the non-semicircular density of states, and at larger RRR the boson-number bands slow down the spreading of the state in Krylov space.