Krylov complexity has it all
Wolfgang Mück · 2026
Read Krylov complexity has it all by Wolfgang Mück (2026) 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-2605-28681.txt.
Introduction
Wolfgang Mück shows that Krylov complexity contains the full information about quantum operator motion. It joins Lanczos coefficients, return amplitude, and spectral density as equal items serving this purpose. A direct looping recipe works out Lanczos numbers from the Taylor series of Krylov complexity around time zero. The paper also clears up why no similar loop exists for spread complexity without extra motion input, offering a proof of principle for it as a complete picture of operator motion.
Catalog record
| CREATOR | Wolfgang Mück |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 13213 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.28681
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Krylov complexity has it all Wolfgang Mück Abstract This paper establishes that Krylov complexity contains the entire information about the dynamics of a quantum operator, extending the list of equivalent quantities that can serve this purpose, such as the Lanczos coefficients, the return amplitude, and the spectral density. To demonstrate this equivalence, an explicit recursive algorithm is constructed to calculate Lanczos coefficients from the Taylor expansion of the Krylov complexity around t = 0 t=0 . Furthermore, the paper discusses the distinction between Krylov and spread complexity, clarifying why a similar recursive algorithm cannot exist for the latter without additional dynamical input. These results provide a “proof of principle” for using Krylov complexity as a complete characterization of operator evolution in quantum systems. Introduction. The definition of Krylov complexity [ 1 ] and the proposal of its use as a measure to characterize chaos in operator dynamics [ 1 , 2 , 3 , 4 , 5 ] has revived interest in the Recursion Method [ 6 ] and related Krylov subspace techniques based on the Lanczos algorithm [ 7 ] . Krylov complexity, in simple terms, is a time-dependent measure for the effective size (dimension) of the subspace of operators that a given observable explores during its time evolution. A similar quantity for quantum states is spread complexity, which measures the spreading of states in Hilbert space during their time evolution [ 8 , 9 , 10 ] . The excellent reviews [ 11 , 12 , 13 ] provide an overview of these recent developments, with applications in fields that range from quantum computing to quantum gravity. It has been known for some time [ 6 ] that the dynamics of an operator \mathcal{O}(t) can be represented in a variety of equivalent ways and that a number of quantities independently contain the entire information about this dynamics. This derives from the following chain of reasoning: In the Heisenberg picture, an operator \mathcal{O} time-evolves as \mathcal{O}(t)=\operatorname{e}^{it\mathcal{L}}\mathcal{O} , where \mathcal{L}=[H,.] is the Liouvillian. Given an operator inner product, (\mathcal{O}|\mathcal{O}^{\prime}) , the Lanczos algorithm diagonalizes the set of operators \mathcal{L}^{n}\mathcal{O} , n=0,1,2\ldots , providing an ordered basis of operators \mathcal{O}_{n} , known as the Krylov basis, which satisfy the recurrence relations \mathcal{L}\mathcal{O}_{n}=b_{n+1}\mathcal{O}_{n+1}+b_{n}\mathcal{O}_{n-1} . The (non-negative) real numbers b_{n} are known as Lanczos coefficients and represent the Liouvillian acting on the Krylov basis. Moreover, defining a chain of wavefunctions \phi_{n}(t)\sim(\mathcal{O}_{n}|\mathcal{O}(t)) , the operator dynamics maps onto a hopping model on this (semi-infinite) chain. Furthermore, the recurrence relations define a set of real orthogonal polynomials with an associated orthogonality measure. Therefore, the following quantities contain, independently, the entire information about the dynamics of \mathcal{O}(t) : (i) the set of Lanczos coefficients, (ii) the return amplitude \phi_{0}(t) or its Laplace transform, (iii) the spectral density associated with the orthogonal polynomials, and (iv) the moments of the spectral density, because they also appear as coefficients in the Taylor expansion of the return amplitude. In this paper I will establish that the above list can be extended by (v) Krylov complexity K(t) . In other words, knowing the Krylov complexity of an operator \mathcal{O}(t) is equivalent to knowing its dynamics. For this purpose, I will explicitly construct an algorithm that recursively calculates the Lanczos coefficients from the coefficients of the Taylor expansion of K(t) around t=0 . I will also discuss why this does not apply to spread complexity and provide a hint on what would be needed in this case. Lanczos coefficients from Krylov complexity. The starting point is the hopping model on the one-dimensional chain associated with the time evolution of \mathcal{O}(t) in the Krylov basis [ 6 , 1 ] , 1 1 1 We follow the convention of [ 14 ] absorbing factors of i into the wavefunctions \phi_{n}(t) such that all of them are real functions. \partial_{t}\phi_{n}(t)=-b_{n+1}\phi_{n+1}(t)+b_{n}\phi_{n-1}(t)~,\qquad\phi_{n}(0)=\delta_{n,0}. (1) Note that b_{n}\geq 0 is ensured by the Lanczos algorithm and b_{n}=0 for n=N terminates it, leading to an N -dimensional Krylov space (and a chain of length N ). Define also \Delta_{n}=b_{n}^{2},\qquad h_{n}=\prod\limits_{k=1}^{n}\Delta_{k},\qquad h_{0}=1~, (2) which naturally appear in the monic version of the Lanczos algorithm [ 14 ] as the Lanczos coefficients ( \Delta_{n} ) and the norm of the n -th Krylov state ( h_{n} ). Then, it is clear that the solutions \phi_{n}(t) of ( 1 ) have Taylor expansions of the form \phi_{n}(t)=\sqrt{h_{n}}\sum_{m=0}^{\infty}C_{n+m,m}\frac{t^{n+2m}}{(n+2m)!}, (3) where the prefactor has been inserted for convenience. The hopping equation ( 1 ) translates into the following recurrence formula for the matrix elements C_{k,l} , C_{k,l}=-\Delta_{k-l+1}C_{k,l-1}+C_{k-1,l}. (4) The initial condition in ( 1 ) implies C_{0,0}=1 , and then ( 4 ) immediately gives C_{k,0}=1\,\forall k . The matrix C_{k,l} is lower-triangular. An important point to remember is that the row C_{p,\cdot} only contains the Lanczos coefficients \Delta_{q} with q\leq p . Consider the Krylov complexity K(t)=\sum_{n=0}^{\infty}n\phi_{n}(t)^{2}, (5) After substituting ( 3 ) and performing a number of summation reorderings, one obtains K(t)=\sum_{p=0}^{\infty}B_{p}\frac{t^{2p}}{(2p)!}, (6) with the coefficients 2 2 2 If, instead of the Krylov complexity, one considers the norm \sum_{n=0}^{\infty}\phi_{n}(t)^{2}=1 , one obtains similar coefficients A_{p} , but without the factor (p-n) . It is a nice exercise to show that ( 4 ) ensures A_{p}=\delta_{p,0} . B_{p}=\sum_{n=0}^{p}\sum_{k=0}^{n}\binom{2p}{p-n+2k}(p-n)h_{p-n}C_{p-n+k,k}C_{p-k,n-k}. (7) The sum on the right-hand side of ( 7 ) is symmetric under k\leftrightarrow n-k . In other words, if one considers the (n,k) -plane, on which the sums in ( 7 ) are represented by the triangle 0\leq n\leq p , 0\leq k\leq n , the sum over the points above the line n=2k is equal to the sum over the points below it. Therefore, ( 7 ) can be expressed as \displaystyle B_{p}=\sum_{k=0}^{[p/2]}\Bigg[ \displaystyle 2\sum_{n=0}^{p-2k-1}\binom{2p}{p-n}(p-n-2k)h_{p-n-2k}C_{p-n-k,k}C_{p-k,n+k} (8) \displaystyle-\binom{2p}{p}(p-2k)h_{p-2k}C_{p-k,k}C_{p-k,k}\Bigg]. The term on the second line compensates the double summation over the points on the line n=2k . By simple inspection one finds that B_{p} contains only the coefficients \Delta_{q} with q\leq p , and that the coefficient \Delta_{p} only appears in the terms with k=0 . Thus, ( 8 ) can be split into B_{p}=B_{p,1}+2\sum_{n=0}^{p-1}\binom{2p}{p-n}(p-n)h_{p-n}C_{p,n}-p\binom{2p}{p}h_{p}, (9) with \displaystyle B_{p,1}=\sum_{k=1}^{[p/2]}\Bigg[ \displaystyle 2\sum_{n=0}^{p-2k-1}\binom{2p}{p-n}(p-n-2k)h_{p-n-2k}C_{p-n-k,k}C_{p-k,n+k} (10) \displaystyle-\binom{2p}{p}(p-2k)h_{p-2k}C_{p-k,k}C_{p-k,k}\Bigg]. The k=0 terms in ( 9 ) can be rewritten, by recursively applying ( 4 ) and using identities and sums for the binomial coefficients, as 2\sum_{n=0}^{p-1}\binom{2p}{p-n}(p-n)h_{p-n}C_{p,n}-p\binom{2p}{p}h_{p}=2\binom{2p-2}{p-1}h_{p}+B_{p,2} (11) with B_{p,2}=4p\sum_{n=0}^{p-2}\binom{2p-2}{p-n-2}h_{p-n-1}C_{p-1,n+1}, (12) which contains only \Delta_{q} with q<p . Thereby, the term with h_{p} has been explicitly isolated in B_{p} . Using the above formulae the algorithm given in Table 1 recursively calculates the Lanczos coefficients \Delta_{p} from the Taylor series coefficients B_{p} . Because the number of input parameters ( N ) is, in principle arbitrary, the Krylov complexity K(t) contains the full information about the entire set of Lanczos coefficients. This does not mean that, given any set of coefficients B_{p} obtained from the Taylor series of some function F(t^{2}) with F(0)=0 , one can find a set of Lanczos coefficients that produces the Krylov complexity K(t)=F(t^{2}) . The caveat is that the input B_{p} is implicitly assumed to really represent a Krylov complexity. Consider the case in which the algorithm stops at some p=P , because \Delta_{P}=0 . In this case, the Krylov space has dimension P , and the remaining input coefficients B_{p} for p>P must uniquely