Abstract
Beginning with the Everett–DeWitt many-worlds interpretation of quantum mechanics, there have been a series of proposals for how the state vector of a quantum system might split at any instant into orthogonal branches, each of which exhibits approximately classical behavior. Here we propose a decomposition of a state vector into branches by finding the minimum of a measure of the mean squared quantum complexity of the branches in the branch decomposition. In a non-relativistic formulation of this proposal, branching occurs repeatedly over time, with each branch splitting successively into further sub-branches among which the branch followed by the real world is chosen randomly according to the Born rule. In a Lorentz covariant version, the real world is a single random draw from the set of branches at asymptotically late time, restored to finite time by sequentially retracing the set of branching events implied by the late time choice. The complexity measure depends on a parameter b with units of volume which sets the boundary between quantum and classical behavior. The value of b is, in principle, accessible to experiment.
Similar content being viewed by others
References
Everett, H.: Relative state formulation of quantum mechanics. Rev. Mod. Phys. 29, 454 (1957)
DeWitt, B.: Quantum mechanics and reality. Phys. Today 23, 30 (1970)
Zeh, H.D.: On the interpretation of measurement in quantum theory. Found. Phys. 1, 69 (1970)
Zurek, W.H.: Pointer basis of quantum mechanics: into what mixture does the wave packet collapse? Phys. Rev. D 24, 1516 (1981)
Zurek, W.H.: Environment-induced superselection rules. Phys. Rev. D 26, 1862 (1982)
Zurek, W.H.: Decoherence, einselection and the quantum origins of the classical. Rev. Mod. Phys. 75, 715 (2003)
Wallace, D.: Everett and structure. Stud. Hist. Philos. Mod. Phys. 34, 87 (2003)
Jess Riedel, C.: Classical branch structure from spatial redundancy in a many-body wave function. Phys. Rev. Lett. 118, 120402 (2017)
Weingarten, D.: Hidden Variable Quantum Mechanics from Branching from Quantum Complexity. (2018). arXiv:1802.10136v5 [quant-ph]
Nielsen, M.A.: A geometric approach to quantum circuit lower bounds. (2005) arXiv:quant-ph/050207
Brown, A.R., Susskind, L.: The second law of quantum complexity. (2018). arXiv:1701.01107 [hep-th]
Kent, A.: Real world interpretations of quantum theory. Found. Phys. 42, 421 (2012)
Kent, A.: A solution to the lorentzian quantum reality problem. Phys. Rev. A 90, 0121027 (2014)
Kent, A.: Quantum reality via late time photodetection. Phys. Rev. A 96, 062121 (2017)
Weingarten, D.: Hidden Variable Theory of a Single World from Many-Worlds Quantum Mechanics. (2017). arXiv:1709.015777v4 [quant-ph]
Page, D.N.: Quantum Uncertainties in the Schmidt Basis Given by Decoherence. (2011). arXiv:1108.2709v2 [hep-th]
Schlosshauer, M.: Decoherence, the measurement problem, and interpretations of quantum mechanics. Rev. Mod. Phys. 76, 1267 (2005)
von Neumann, J.: Mathematical Foundations of Quantum Mechanics. Princeton University, Princeton (1955)
Eugene, P.W.: Remarks on the mind-body problem. In: Good, I.J. (ed.) The Scientist Speculates. Heineman, London (1961)
London, F., Bauer, E.: La théorie de l’observation en méchanique quantique. In: Wheeler, J., Zurek, W.H. (eds.) Quantum Theory, Measurement. Princeton University, Princeton (1983)
DiVincenzo, D.P.: Two-bit gates are universal for quantum computation. Phys. Rev. A 51, 1015 (1995)
Acknowledgements
Thanks to Jess Riedel for an extended debate over an earlier version of this work and to an anonymous reviewer for comments leading to many improvements incorporated in the present version.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix 1: Truncated Hermitian Operator Hilbert Space
Let \({\mathcal {H}}_x^n\) be the subspace of \({\mathcal {H}}_x\) with less than n bosons. The dimension \(d_n\) of each \({\mathcal {H}}_x^n\) is finite. Let \({\mathcal {H}}^n\) be the product over x of all \({\mathcal {H}}^n_x\)
For any site x, let \({\mathcal {F}}^n_x\) consist of all Hermitian \(f_x\) on \({\mathcal {H}}^n_x\) with finite
and vanishing trace
For any pair of nearest neighbor sites \(\{x, y\}\), let \({\mathcal {F}}^n_{xy}\) consist of all Hermitian \(f_{xy}\) on \({\mathcal {H}}^n_x \otimes {\mathcal {H}}^n_y\) with finite
and vanishing traces
Inner products on \({\mathcal {F}}^n_x\) and \({\mathcal {F}}^n_{xy}\) are
Operators \(f_x \in {\mathcal {F}}^n_x\) and \(f_{xy} \in {\mathcal {F}}^n_{xy}\) can be made into operators on \({\mathcal {H}}^n\) by
where \(I_q\) is the identity operator on \({\mathcal {H}}^n_q\). As usual, we now drop the hat and use the same symbol for operators on \({\mathcal {H}}^n_x\), \({\mathcal {H}}^n_x \otimes {\mathcal {H}}^n_y\), and the corresponding operators on \({\mathcal {H}}^n\).
Let \(K^n\) be the vector space over the reals of linear operators k on \({\mathcal {H}}^n\) given by sums of the form
for any collection of \(f_{x y} \in {\mathcal {F}}^n_{x y}\) for a set of nearest neighbor pairs \(\{x, y\}\) and any collection of \(f_x \in {\mathcal {F}}^n_x\) in a set of sites x. The inner product on K is
An equivalent inner product on \(K^n\), which is a version of the inner product on operator Hilbert space in [10], is
where \(\mathrm {Tr}\) is the trace on all of \({\mathcal {H}}^n\) and \(n_L\) is the number of sites in the lattice L. As a result of the factor of \(\frac{1}{\sqrt{d_n}}\) in Eq. (A8), if \(d_n\) is made large, matrix elements of k given by Eq. (A8) will approach those of k given by Eq. (15) and \(K^n\) will become equivalent to the operator space K of Sect. 3.2.
Appendix 2: Lower Bound on the Complexity of Entangled States
The proof of Eq. (37) proceeds as follows. The trajectories \(k(\nu ) \in K\) and \(U_k(\nu )\) which determine any \(C( |\psi \rangle , |\omega \rangle )\), according to Eqs. (17a)–(20), we characterize by a corresponding set of trajectories of Schmidt spectrum vectors. We then find the rotation matrices which govern the motion of these vectors as \(\nu\) changes. A bound on the time integral of the angles which occur in these matrices by a time integral of \(\parallel k(\nu ) \parallel\) yields Eq. (37).
1.1 2.1 Schmidt Spectra
Consider some entangled n-fermion \(|\psi \rangle\) of form Eq. (36). For a trajectory \(k(\nu ) \in K\), let \(U_k(\nu )\) be the solution to Eqs. (17a) and (17b). Define \(|\omega (\nu ) \rangle\) to be
for some product state \(|\omega \rangle\) and assume that \(k(\nu )\) has been chosen to give
for a phase factor \(\xi\). Since all \(k(\nu )\) conserve fermion number, \(|\omega \rangle\) according to Eq. (9) must have the form
for some number of bosons m.
We now divide the lattice L into a collection of disjoint regions and define a corresponding collection of Schmidt decompositions of the trajectory of states which determine any \(C( |\psi \rangle , |\omega \rangle )\). Divide L into subsets \(L^e, L^o\), with, respectively, even or odd values of the sums of components \({\hat{x}}_i\). The sites in each subset have nearest neighbors only in the other. Let \(D^e_{ij}, D^o_{ij}, D^e, D^o\) be
Between \(D^e\) and \(D^o\) choose the larger, or either if they are equal. Assume the set chosen is \(D^e\). Among the nm spins \(s_{ij}\), at least \(\frac{nm}{2}\) will have the same value and therefore correspond to \(D_{ij}\) which do not intersect. The corresponding collection of \(D^e_{ij}\) will then include at least \(\frac{nmV}{4}\) points.
From this set of \(D^e_{ij}\) construct a set of subsets \(E_\ell\) each consisting of 2n distinct points chosen from 2n distinct \(D^e_{ij}\). The total number of \(E_\ell\) will then be at least \(\frac{m V}{8}\). We will consider only the first \(\frac{m V}{8}\) of these.
The Hilbert space \({\mathcal {H}}\) is given by a tensor product
of a fermion space \({\mathcal {H}}^f\) and a boson space \({\mathcal {H}}^b\). Similarly the space \({\mathcal {H}}_x\) at each x is given by a tensor product
of a fermion space \({\mathcal {H}}_x^f\) and a boson space \({\mathcal {H}}_x^b\). The dimensions of \({\mathcal {H}}_x^f\) and \({\mathcal {H}}_x^b\) are, respectively, 4 and \(\infty\).
For each set \(E_\ell\) form the tensor product spaces
It follows that \({\mathcal {Q}}_\ell\) has dimension \(4^{2n}\) and
A Schmidt decomposition of \(|\omega (\nu ) \rangle\) according to Eq. (B8) then becomes
where
for \(0 \le j < 4^{2n}\) and real non-negative \(\lambda _{j\ell }( \nu )\) which fulfill the normalization condition
Each \(|\phi _{j\ell }(\nu ) \rangle\) is a pure fermion state while the \(|\chi _{j\ell }(\nu ) \rangle\) can include both fermions and bosons.
The fermion number operators \(N[{\mathcal {Q}}_\ell ]\) and \(N[{\mathcal {R}}_\ell ]\) commute and \(|\omega (\nu ) \rangle\) is an eigenvector of the sum with eigenvalue n. It follows that the decomposition of Eq. (B9) can be done with \(|\phi _{j\ell }( \nu ) \rangle\) and \(|\chi _{j\ell }(\nu ) \rangle\) eigenvectors of \(N[{\mathcal {Q}}_\ell ]\) and \(N[{\mathcal {R}}_\ell ]\), respectively, with eigenvalues summing to n. Let \(|\phi _{0\ell } \rangle\) be \(|\Omega _\ell \rangle\), the vacuum state of \({\mathcal {Q}}_\ell\), and let \(|\phi _{i\ell } (\nu ) \rangle , 1 \le i \le 4n\), span the 4n-dimensional subspace of \({\mathcal {Q}}_\ell\) with \(N[{\mathcal {Q}}_\ell ]\) of 1. We assume the corresponding \(\lambda _{i\ell }( \nu ), 1 \le i \le 4n\), are in nonincreasing order. Consider Eq. (B9) for \(\nu = 1\). By Eq. (B2), for any choice of \(\ell\) there will be a set of 2n nonzero orthogonal \(|\phi _{1\ell }( 1) \rangle , \ldots |\phi _{2n\ell }( 1) \rangle\) with
for \(1 \le j \le 2n\).
On the other hand, for \(\nu = 0\), Eq. (B9) becomes a decomposition of the product state \(|\omega \rangle\). The boson part of \(|\omega (0) \rangle\) will occur as the same overall tensor factor in each \(|\chi _{1\ell }(0) \rangle ,\ldots |\chi _{n\ell }(0) \rangle\). The fermion part of \(|\omega (0) \rangle\) is a product of n independent single fermion states, the space spanned by the projection of these into some \({\mathcal {Q}}_\ell\) is at most n dimensional, and as a result at most n orthogonal \(|\phi _{1\ell }(0) \rangle ,\ldots |\phi _{n\ell }(0) \rangle\) can occur. Therefore at \(\nu = 0\), there will be at most n nonzero \(\lambda _{1\ell }(0), \ldots \lambda _{n\ell }(0)\). For \(n < j \le 2n\), we have
But according to Eq. (B11), for each fixed value of \(\ell\) the set of components \(\{\lambda _{j\ell }( \nu )\}\) indexed by j is a unit vector. Eqs. (B13) and (B12) then imply that as \(\nu\) goes from 0 to 1, \(\{\lambda _{j\ell }( \nu )\}\) must rotate through a total angle of at least \(\arcsin (\sqrt{\frac{n}{mV}})\).
For the small interval from \(\nu\) to \(\nu + \delta \nu\) let \(\mu _{j\ell }(\nu )\) and \(\theta _{\ell }(\nu )\) be
We then have
Summed over the \(\frac{mV}{8}\) values of \(\ell\), Eq. (B15) becomes
and therefore
1.2 2.2 More Schmidt Spectra
Replacing the subsets \(E_\ell\) defined in Appendix B1, with subsets of L obtained from the \(S_\ell\) of Sect. 4 leads to an additional bound similar to Eq. (B17).
For each \(0 \le \ell < q\), of the two subsets of L defined by \(S_\ell\), let \(T_\ell\) be the subset which, for each \(0 \le i < m\), holds \(n_0\) of the sets \(D_{ij}, 0 \le j < n\). Redefine \({\mathcal {Q}}_\ell , {\mathcal {R}}_\ell\) of Eqs. (B7a) and (B7b), to be
For each \(0 \le \ell < q\) there is again a corresponding Schmidt decomposition of \(|\omega (\nu ) \rangle\) of Eqs. (B1) and (B2)
where
Each \(|\phi ^T_{j\ell }(\nu ) \rangle\) is a pure fermion state while the \(|\chi ^T_{j\ell }(\nu ) \rangle\) can include both fermions and bosons. For \(\nu = 1\), for every \(0 \le \ell < q\), the sum over j in Eq. (B19) has m nonzero entries each with
with \(|\phi ^T_{j\ell }(1) \rangle\) carrying fermion number \(n_0\) and \(|\chi ^T_{j\ell }(1) \rangle\) carrying fermion number \(n_1\).
Duplicating the discussion of Appendix B1, a trajectory of angles \(\theta ^T_\ell (\nu )\) can be defined which rotates the unit vector \([ \lambda ^T_{j\ell }(0) ]\) arising from the product state \(|\omega (0) \rangle\) into the unit vector \([\lambda ^T_{j\ell }(1) ]\) of Eq. (B21). For each \(0 \le \ell < q\), a version of the lower bound of Eq. (B15) can be obtained by finding the product state \(|\omega (0) \rangle\) which gives \([ \lambda ^T_{j\ell }(0) ]\) closest to \([ \lambda ^T_{j\ell }(1) ]\) for the set of \(0 \le j < m\) corresponding to \(|\phi ^T_{j\ell }(0) \rangle\) and \(|\chi ^T_{j\ell }(0) \rangle\) with fermion numbers \(n_0\) and \(n_1\), respectively.
According to Eq. (9), the product state \(|\omega (0) \rangle\) includes n fermion creation operators \(d^\dagger _f( p_i)\) given by Eq. (8a). Since \(|\omega (1) \rangle\) and therefore \(|\omega (0) \rangle\) are normalized to 1, we can require the \(p_i( x, s)\) to be orthonormal. The simplest way to insure \(n_0\) and \(n_1\), respectively, for \(|\phi _{0\ell }(0) \rangle\) and \(|\chi _{0\ell }(0) \rangle\) is for the support of \(p_i( x, s)\) to be entirely within \(T_\ell\) for \(0 \le i < n_0\) and entirely outside \(T_\ell\) for \(n_0 \le i < n\). The Schmidt decomposition of Eq. (B19) then yields a vector \([ \lambda ^T_{j\ell }(0) ]\) with only a single nonzero entry and therefore
A larger value of the sum in Eq. (B22) is possible only if an even number of \(p_i(x,s)\) have support both within \(T_\ell\) and outside \(T_\ell\). For some \(r \le n_0, n_1,\), define z to be the set
Then for \(i \in z\), suppose
where the \(p^0_i( x, s)\) have support entirely within \(T_\ell\) and the \(p^1_i( x, s)\) have support entirely outside \(T_\ell\). Since \(p_i(x,s)\) is normalized and the support of \(p^0_i(x,s)\) is disjoint from the support of \(p^1_i(x,s)\) we have
The piece \(|{\hat{\omega }}( 0) \rangle\) of \(|\omega ( 0) \rangle\) with fermion number \(n_0\) on \(T_\ell\) and \(n_1\) outside \(T_\ell\) is given by
where the sum is over all r element subsets \(u \subset z\).
The vector \([ \lambda ^T_{j\ell }(0) ]\) corresponding to \(|{\hat{\omega }}(0) \rangle\) will have at most \(\frac{(2r)!}{(r!)^2}\) nonzero entries, one for each of the sets u in the sum in Eq. (B26). The Cauchy-Schwartz inequality then yields
Let \(|{\hat{p}}^0_i \rangle\) be the projection of \(|p^0_i \rangle\) orthogonal to all \(|{\hat{p}}^0_j \rangle , j < i\), and let \(|{\hat{p}}^1_i \rangle\) be the projection of \(|p^1_i \rangle\) orthogonal to all \(|{\hat{p}}^1_j \rangle , j < i\). Substituting \(\{ |{\hat{p}}^0_i \rangle \}\) and \(\{ |{\hat{p}}^1_i \rangle \}\) for \(\{|p^0_i \rangle \}\) and \(\{ |p^1_i \rangle \}\), respectively, in Eq. (B26) leaves \(|{\hat{\omega }}(0) \rangle\) unchanged. The value of \(\parallel |{\hat{\omega }}(0) \rangle \parallel\) will then be maximized if the resulting \(\parallel | {\hat{p}}^0_i \rangle \parallel\) and \(\parallel | {\hat{p}}^1_i \rangle \parallel\) are increased as needed to satisfy Eq. (B25).
Suppose now that \(\parallel |{\hat{\omega }}(0) \rangle \parallel\) has been maximized with respect to \(\parallel | {\hat{p}}^0_i \rangle \parallel\) and \(\parallel | {\hat{p}}^1_i \rangle \parallel\) for all \(0 \le i < r\), expect some pair of values j, k. The remaining dependence on \(\parallel |{\hat{p}}^0_i \rangle \parallel\) and \(\parallel |{\hat{p}}^1_i \parallel\) for \(i = j, k,\) is maximized at
If \(\parallel |{\hat{\omega }}(0) \rangle \parallel\) is then maximized with respect to the remaining i independent \(\parallel |{\hat{p}}^0_i \rangle \parallel\) and \(\parallel | {\hat{p}}^1_i \rangle \parallel\), Eq. (B27) becomes
Suppose m has the form \(\frac{(2r)!}{(r!)^2}\). For any \(r' \le r\) Eq. (B29) becomes
An induction argument then shows that Eq. (B30) is an increasing function of \(r'\). For \(r' > r\) on the other hand, the sum in Eq. (B30) becomes
which is a decreasing function of \(r'\). The maximum of Eq. (B30) will therefore be at \(r' = r\).
Now suppose m lies between \(\frac{(2r)!}{(r!)^2}\) and \(\frac{(2r + 2)!}{[(r+1)!]^2}\). For \(r' \le r\) the maximum Eq. (B30) will still be at \(r' = r\) and given by
For \(r' = r + 1\), Eq. (B30) becomes
A further induction argument shows that Eqs. (B32b) and (B33b) are decreasing functions of r. Thus for \(m \ge 2\), we have
A duplicate of the argument leading to Eq. (B17) then yields
1.3 2.3 Schmidt Rotation Matrix
A lower bound on \(C(|\psi \rangle )\) follows from Eqs. (B17) and (B35). Appendices 2.3 and 2.4 derive the consequence of Eq. (B17). A derivation of the additional terms in the bound on \(C(|\psi \rangle )\) which follow from Eq. (B35) is briefly summarized in Appendix 2.5.
The rotation of \(\lambda _{j\ell } (\nu )\) during the interval from \(\nu\) to \(\nu + \delta \nu\) will be determined by \(k(\nu )\). For each \(f_{xy}\) in Eq. (15) for \(k(\nu )\) which can contribute to a nonzero value of \(\theta _{\ell }(\nu )\), the nearest neighbor pair \(\{x, y\}\) has one point, say x in \(E_{\ell }\). Since \(E_{\ell } \subset D^e\) and the nearest neighbors of all points in \(D^e\) are in \(D^o\), y can not be in \(E_\ell\). Let \(g_{\ell }(\nu )\) be the sum of all such \(f_{xy}\). The effect of all other terms in Eq. (15) on the Schmidt decomposition of Eq. (B9) will be a unitary transformation on \({\mathcal {R}}_\ell\) and identity on \({\mathcal {Q}}_\ell\). All other terms will therefore leave \(\lambda _{j\ell }( \nu )\) unchanged.
The effect of \(g_{\ell }(\nu )\) on \(\lambda _{j\ell }(\nu )\) over the interval from \(\nu\) to \(\nu + \delta \nu\) can be determined from the simplification
From \(|\omega (\nu + \delta \nu ) \rangle \langle \omega (\nu + \delta \nu )|\) of Eq. (B36), construct the density matrix \(\rho (\nu + \delta \nu )\) by a partial trace over \({\mathcal {R}}_{\ell }\), using the basis for \({\mathcal {R}}_{\ell }\) from the Schmidt decomposition in Eq. (B9) of \(|\omega (\nu ) \rangle\) at t
An eigenvector decomposition of \(\rho (\nu + \delta \nu )\) exposes the \(\lambda _{j\ell }(\nu + \delta \nu )\)
A power series expansion through first order in \(\delta \nu\) applied to Eqs. (B36), (B37) and (B38) then gives for \(\mu _{j\ell }(\nu )\) of Eq. (B14a)
for the rotation matrix \(r_{jk\ell }(\nu )\)
1.4 2.4 Rotation Angle Bounds
Since the \(f_{xy}\) contributing to \(g_\ell (\nu )\) conserve total fermion number N, \(g_\ell (\nu )\) can be expanded as
where \(z^0( x, y, \nu )\) acts only on states with \(N( {\mathcal {H}}_x \otimes {\mathcal {H}}_y)\) of 0, \(z^1( x, y, \nu )\) acts only on states with \(N( {\mathcal {H}}_x \otimes {\mathcal {H}}_y)\) strictly greater than 0, and the \(z^i(x,y,\nu )\) are normalized by
The operator \(z^0(x, y, \nu )\) will be
where \(P^f(x,y)\) projects onto the vacuum state of \({\mathcal {H}}^f_x \otimes {\mathcal {H}}^f_y\) and \(g^b(x,y,\nu )\) is a normalized Hermitian operator acting on \({\mathcal {H}}^b_x \otimes {\mathcal {H}}^b_y\)
Combining Eqs. (B14b),(B39) - (B41b) gives
with the definitions
Since the \(|\phi _{j\ell }(\nu ) \rangle\) are orthonormal, \(g^b(x,y,\nu )\) is Hermitian and the \(\lambda _{k\ell }(\nu )\) are real, we have
Equation (B45) for \(i = 0\) can then be turned into
But in addition
Also \(I - z^{0f}(x,y)\) is a projection operator so that
The normalization condition on \(z^0(x,y,\nu )\) implies \([g^b(x,y,\nu )]^2\) has trace 1 as an operator on \({\mathcal {H}}^b_x \otimes {\mathcal {H}}^b_y\) and therefore all eigenvalues bounded by 1. Eqs. (B43a), (B44b), (B47), (B48), and (B49) then give
For \(\mu ^1_{j\ell }(x,y,\nu )\), since \(z^1(x,y,\nu )\) is nonzero only on the subspace with \(N({\mathcal {H}}_x \otimes {\mathcal {H}}_y)\) nonzero, we have
Equations (B44b) and (B51) give
But by Eq. (B42), \([z^1(x,y,\nu )]^2\) as an operator on \({\mathcal {H}}_x \otimes {\mathcal {H}}_y\), has trace 1 and therefore all eigenvalues bounded by 1. Thus Eq. (B52) implies
By construction of \(D^e\), each nearest neighbor pair \(\{x, y\}\) with \(x \in D^e\) must have \(y \in D^o\). Also any \(x \in D^e\) is contained in at most a single \(E_\ell\). As a result Eqs. (B44a), (B50) and (B53) imply
The Cauchy–Schwartz inequality then gives
The state \(|\omega (\nu ) \rangle\) can be expanded as a linear combination of orthogonal states each with n fermions each at a single position. A state with fermions at n positions will survive the projection \(I - z^{0f}(x,y)\) only if at least one of the fermions is either at x or y. Each \(x \in D^e\) can be the member of only a single such pair of nearest neighbor \(\{x, y\}\). A \(y \in D^o\) can be in 6 x, y pairs for an \(x \in D^e\). Thus a term with n fermion positions in the expansion of \(|\omega (\nu ) \rangle\) will pass \(I - z^{0f}(x,y)\) for at most 6n pairs of x and y. Therefore
By Eq. (16)
In addition, \(z^0(x,y,\nu )\) is orthogonal to \(z^1(x, y, \nu )\). It follows that
Assembling Eqs. (B55), (B56) and (B58) gives
Eq. (B17) then implies
and therefore
Since Eq. (B61) holds for all product \(|\omega \rangle\) we obtain
1.5 2.5 Additional Terms
The nearest neighbor \(\{x,y\}\) which contribute to each \(\theta ^T_{\ell }(\nu )\) in Eq. (B35) are all distinct from the pairs which contribute to \(\theta _{\ell }(\nu )\) in Eq. (B17). A repeat of the steps leading to Eq. (B59) yields
Eq. (B62) becomes
Appendix 3: Upper Bound on the Complexity of Entangled States
An upper bound on \(C( |\psi \rangle )\) of the n-particle entangled state of Eq. (36) is given by \(C( |\psi \rangle , |\omega \rangle )\) for any n-particle product state \(|\omega \rangle\), for which in turn an upper bound is given by
for any trajectory \(k(\nu ) \in K\) fulfilling Eqs. (B1) and (B2). Beginning with an \(|\omega \rangle\) consisting of n particles each at one of a corresponding set of n single points, we construct a sufficient \(k(\nu )\) in three stages. First,\(|\omega \rangle\) is split into a sum of m orthogonal product states, each again consisting of n particles one at each of a corresponding set of n single points. Then the position of each of the particles in the product states is moved to the center of the wave function of one of the single particle states of Eq. (35). Finally, by approximately \(\ln ( V) / \ln ( 8)\) iterations of a fan-out tree, the mn wave functions concentrated at points are spread over the mn cubes \(D_{ij}\).
1.1 3.1 Product State to Entangled State
Define the set of positions \(x_{ij}\) to be
for \(0 \le i< m, 0 \le j < n\) and arbitrary base point \(x_{00}\). Let the set of n-particle product states \(|\omega _i \rangle\) be
The entangle n-particle state \(|\chi \rangle\)
we generate from a sequence of unitary transforms acting on \(|\omega \rangle = |\omega _0 \rangle\).
Let \(k_{0}\) acting on \({\mathcal {H}}_{x_{00}} \otimes {\mathcal {H}}_{x_{01}}\) have matrix elements
and extend \(k_0\) to \({\mathcal {H}}\) by Eq. (14). We then have
where
and the set of spin indices \(s_{ij}, 0 \le i,j < n\) is
Now let \(k_{1}\) acting on \({\mathcal {H}}_{x_{01}} \otimes {\mathcal {H}}_{x_{02}}\) have matrix elements
and extend \(k_1\) to \({\mathcal {H}}\) by Eq. (14). We then have
for \(\theta _1\) given by \(\frac{\pi }{2}\).
Continuing in analogy to Eqs. (C5)–(C12), for a sequence of operators \(k_j\), \(0 \le j < n-1\), acting on \({\mathcal {H}}_{x_{0j}} \otimes {\mathcal {H}}_{x_{0j+1}}\), and corresponding \(\theta _j\) we obtain
Let \(k_{n-1}\) acting on \({\mathcal {H}}_{x_{00}} \otimes {\mathcal {H}}_{x_{10}}\) have matrix elements
extend \(k_{n-1}\) to \({\mathcal {H}}\) by Eq. (14), and let \(\theta _{n-1}\) be \(\frac{\pi }{2}\). Applying \(\exp (i \theta _{n-1} k_{n-1})\) to Eq. (C13), followed by a similar sequence of \(\exp (i \theta _j k_j), n \le j < 2n -1\) acting on \({\mathcal {H}}_{x_{0(j-n+1)}} \otimes {\mathcal {H}}_{x_{1(j-n+1)}}\) gives
Multiplying Eq. (C15) by \(\exp (i\theta _jk_j), 2n-1 \le j < 3n-2\) on \({\mathcal {H}}_{x_{1(j-2n+1)}} \otimes {\mathcal {H}}_{x_{1(j-2n +2)}}\), and then \(\exp (i\theta _jk_j), 3n -2 \le j < 4n -2\) on \({\mathcal {H}}_{x_{1(j-3n+2)}} \otimes {\mathcal {H}}_{x_{2(j-3n+2)}}\) gives
The end result of a sequence of \(2mn - m\) such steps is \(|\chi \rangle\) of Eq. (C4)
The \(k_i\) and \(\theta _i\) of Eq. (C17) have
Thus Eq. (C17) implies
1.2 3.2 Entangled State Repositioned
Let \(y_{ij}\) be the center of cube \(D_{ij}\) of Eq. (35), \(s_{ij}\) the spins of Eq. (35) and \(\zeta _i\) the phases of Eq. (36). Define the entangled n-particle state \(|\phi \rangle\) be
For each \(0 \le i< m, 0 \le j < n\), let \(z^0_{ij}, z^1_{ij} \ldots z^{r_{ij}}_{ij}\) be the shortest sequence of nearest neighbor sites such that
for the \(x_{ij}\) in Eqs. (C2a)–(C4) and such that all \(z^\ell _{ij}\) for distinct \(\ell , i, j,\) are themselves distinct. For each \(0 \le \ell < r_{ij} -1\), for nearest neighbor pair \(z^\ell _{ij}, z^{\ell +1}_{ij}\), let \(k^\ell _{ij}\) acting on \({\mathcal {H}}_{z^\ell _{ij}} \otimes {\mathcal {H}}_{z^{\ell +1}_{ij}}\) have matrix elements
and extend \(k^\ell _{ij}\) to \({\mathcal {H}}\) by Eq. (14). For each i, j pair with \(j<n-1\), for the final nearest neighbor step \(\exp ( i k^\ell _{ij}), \ell = r_{ij} - 1,\) Eqs. (C22a) and (C22b) are modified to produce spin orientation \(s_{ij}\) at \(y_{ij}\)
and for \(j = n-1\) for the final \(\exp ( i k^\ell _{in-1}), \ell = r_{in-1} - 1,\) Eqs. (C22a) and (C22b) are modified in addition to generate the phase \(\zeta _i\)
Define r to be
and for each i, j pair define
Let \(k^\ell\) be
Then we have
for \(|\chi \rangle\) of Eq. (C4).
The \(k^\ell\) of Eqs. (C27), (C22a) - (C24b) have
Thus Eq. (C28) implies
We now minimize r over the base point \(x_{00}\)
with the result
where we have dropped the hat on r.
1.3 3.3 Fan-Out
The state \(|\phi \rangle\) with particles at the centers of the cubes \(D_{ij}\) we now fan-out to the state \(|\psi \rangle\) with particle wave functions spread uniformly over the cubes \(D_{ij}\). For sufficiently small lattice spacing a nearly all of the complexity of the bound on \(C(|\psi \rangle )\) is generated in this step.
Let d be the length of the edge of the \(D_{ij}\). Each edge of \(D_{ij}\) then consists of \(d+1\) sites. The volume V is then \(d^3\). We begin with case
for some integer p. For simplicity we present the fan-out applied to a prototype single particle state \(|\upsilon _0 \rangle\) on prototype cube G with edge length d, and center at some point y
The first stage of the fan-out process consists of splitting \(|\upsilon _0 \rangle\) into a pair of components displaced from each other in lattice direction 1. For integer \(-2^{p-2} \le i \le 2^{p-2}\) define y(i) to be y incremented by i nearest neighbor steps in lattice direction 1. For \(1 \le j \le 2^{p-2}\) define \(k_j\) on \({\mathcal {H}}_{y( j -1)} \otimes {\mathcal {H}}_{y(j)}\) to have matrix elements
For \(-2^{p-2} \le j \le -1\) define \(k_j\) by Eqs. (C35a) and (C35b) but with \(j+1\) in place of \(j-1\). Then define \({\bar{k}}_j\) by
With these definitions it then follows that
for \(m = 2^{p-2}\), is given by
Equations (C36a) and (C36b) imply
It then follows that
where for simplicity we have used an overestimate for \(\parallel {\bar{k}}_1 \parallel\).
The next stage of the fan-out consists of splitting each of the 2 components of \(|\upsilon _1 \rangle\) but now in lattice direction 2. For \({\bar{k}}_j, 2^{p-2} < j \le 2^{p-1}\), defined by adapting of Eqs. (C35a)–(C36b), we have
with \(m = 2^{p-1}\), given by
for y(i, j) defined to be y(i) displaced j steps in lattice direction 2. Eqs. (C36a) and (C36b) adapted to the fan-out in direction 2 give \({\bar{k}}_j, 2^{p-2} < j \le 2^{p-1}\) each acting on twice as many sites as was the case for the direction 1 fan-out and therefore
It then follows that
Splitting yet again, now in lattice direction 3, yields
for \(m =2^{p-1} + 2^{p-2}\), given by
for \(y(i,j, \ell )\) defined to be y(i, j) displaced \(\ell\) steps in lattice direction 3.
Eqs. (C36a) and (C36b) adapted to the fan-out in direction 3 give \({\bar{k}}_j, 2^{p-1} < j \le 2^{p-1} + 2^{p-2}\), each acting on twice as many sites as was the case for the direction 2 fan-out and therefore
It then follows that
The weight originally concentrated in \(|\upsilon _0 \rangle\) at the center point y of G, with edge length d, in \(|\upsilon _3 \rangle\) is distributed equally over the center points of 8 sub-cubes of G each with edge length \(\frac{d}{2}\). Combining Eqs. (C40), (C44) and (C48) gives
The fan-out process of Eqs. (C37)–(C49) we now repeat a total of \(p-1\) iterations arriving at a state \(|\upsilon _{3 p - 3} \rangle\) with weight equally distributed over the center points of \(2^{3 p - 3}\) cubes each with edge length 2. Eqs. (C49) rescaled for iteration \(\ell\) give
The term \(2^{p-\ell - 1}\) counts the decreasing number of lattice steps between cube centers as the fan-out process is iterated, while the term \(2^{\frac{3\ell - 3}{2}}\) counts the growing number of cubes and therefore of sites which each subsequent operator \({\bar{k}}(i)\) acts on simultaneously.
To complete the fan-out process, the weight at the center of each of the cubes with edge length 2 needs to be distributed to the 26 points forming its boundary. This process can be carried out in 3 additional steps thereby defining \(|\upsilon _{3p -2} \rangle , |\upsilon _{3p-1} \rangle\) and \(|\upsilon _{3p} \rangle\).
To obtain \(|\upsilon _{3p -2} \rangle\) from \(|\upsilon _{3p -3} \rangle\), the weight at the center of each edge length 2 cube is distributed simultaneously and equally to the points at the centers of the 6 edge length 2 squares forming the cube’s boundary. This process itself is done simultaneously across all \(2^{3 p - 3}\) cubes. The result is
To obtain \(|\upsilon _{3p -1} \rangle\) from \(|\upsilon _{3p -2} \rangle\), the weight at the center of each edge length 2 square is distributed simultaneously and equally to the center point of the 4 length 2 lines forming the boundary of that square. This process itself is done simultaneously across all faces of all \(2^{3 p - 3}\) cubes. The result is
To obtain \(|\upsilon _{3p} \rangle\) from \(|\upsilon _{3p -1} \rangle\), the weight at the center of each length 2 line is distributed simultaneously and equally to that line’s pair of end points. This process itself is done simultaneously across all lines forming the boundaries of the faces of all \(2^{3 p - 3}\) cubes. The result is
The bound on \(C( |\upsilon _{3p} \rangle , |\upsilon _{3 p - 3} \rangle )\) obtained by summing Eqs. (C51)–(C53) turns out to be less than the bound in Eq. (C50) for \(\ell = p\). We therefore sum Eq. (C50) from \(\ell\) of 1 to p and obtain
Substituting V for \(2^{3p}\), we then have
The bound of Eq. (C54) is derived assuming Eq. (C33) giving the edge d of cube G as an even power of 2. Consider now the case
Assume again that at each iteration \(\ell\) of the fan-out process, each edge length of each parent cube is split as evenly as possible into halves to produce 8 child cubes with all edges nearly equal. Suppose d is \(2^p - 1\). After iteration \(\ell\) has been completed, the total number of cubes will still be \(2^{3 \ell }\). Orthogonal to each direction, the cubes can be grouped into \(2^\ell\) planes, each holding \(2^{2 \ell }\) cubes. But for each direction one of these orthogonal planes will have an edge in that direction which is one lattice unit shorter than the corresponding edge of the other \(2^\ell\) planes. It follows that the update process in each direction can proceed with \(2^{p - \ell - 1} - 1\) steps occuring simultaneously across all cubes, and one final update skipped for the cubes with a single edge in that direction one lattice unit shorter. The bound of Eq. (C50) will hold without modification. For d given by \(2^p - 2\), after iteration \(\ell\), for each direction, there will be two planes of \(2^{2 \ell }\) cubes each with the edge in that direction one lattice unit shorter. The bound of Eq. (C50) will continue to hold. Similarly for d given by \(2^p - q\) for any \(q < 2^{p-1}\).
For d of Eq. (C56), when \(\ell\) reaches \(p - 1\) the resulting cubes (no longer exactly cubes) will have a mix of edges of length 2 and of length 1. The argument leading to Eqs. (C51)–(C53) can be adapted to show they continue to hold for the final pass with \(\ell\) of p. The bound of Eq. (C54) remains in place for the net result of the entire fan-out process. By assumption, according to Eq. (C56) we have
Then since V is \(d^3\), Eq. (C54) gives
which is weaker than Eq. (C55) and therefore holds whether or not d is an even power of 2.
The bound of Eq. (C55) applies to a fan-out process on a single prototype state on cube G. Assume the process repeated in parallel on the mn cubes \(D_{ij}\), thereby generating \(|\psi \rangle\) of Eq. (36). For \(|\phi \rangle\) of Eq. (C20) we then have
From Eqs. (C19) and (C32), it follows that for a product state \(|\omega \rangle\) we have
where
for r of Eq. (C32). Eq. (38) then follows.
Appendix 4: Complexity Group
We now show that the topological closure of the group G of all \(U_k( 1)\) realizable as solutions to Eqs. (17a) and (17b) has as a subgroup the direct product
where \(SU(d_n)\) acts on the subspace of \({\mathcal {H}}\) with eigenvalue n of the fermion number operator N, \(d_n\) is the dimension of this subspace, and the product is over the range \(0 \le n \le 16 B^3\).
1.1 4.1 Lie Algebras
The \(8 B^3\) sites of the lattice L we reorder as a 1-dimensional array of distinct sites, successive pairs of which are nearest neighbors with respect to the original lattice L. The new array of sites we label with an integer valued index z ranging from 0 to \(8 B^3 -1\).
For any pair of nearest neighbor \(\{z, z'\}\), let \({\mathcal {F}}_{z z'}\) be the set of operators of the form
where \(I_q\) is the identity operator on \({\mathcal {H}}_q\) and \(g_{zz'}\) is a traceless Hermitian operator acting on \({\mathcal {H}}_z \otimes {\mathcal {H}}_{z'}\) which commutes with \(N_{zz'}\), the fermion number operator on \({\mathcal {H}}_z \otimes {\mathcal {H}}_{z'}\). Let \(K_p\) be the vector space over the reals of operators of the form
for any collection of \(f_{z z'} \in {\mathcal {F}}_{zz'}\) for \(z, z' \le p\).
Let \(G_p\) be the group on \({\mathcal {H}}\) of all \(U_k(1)\) realizable as solutions to Eq. (17a) for \(k(\nu ) \in K_p\). The topological closure of the group \(G_p\) consists of all operators of the form \(\exp ( i h)\) for \(h \in L_p\), where \(L_p\) is the Lie algebra generated by \(K_p\) [21]. Said differently, \(L_p\) is the smallest set of operators such that \(K_p \subseteq L_p\) and, in addition, for any \(h_0, h_1 \in L_p\), and any real \(r_0, r_1\), there are \(h_2, h_3 \in L_p\) given by
The requirement that \(L_p\) be closed under sums in Eq. (D4a) follows from the Trotter product formula applied to the large t limit
The requirement that \(L_p\) be closed under commutation in Eq. (D4b) follows from the Baker–Campbell–Hausdorff formula applied to the large t limit
The requirement of taking a topological closure of the group generated by \(U_k(1)\) in order to generate \(L_p\) is a consequence of the appearance of limits in Eqs. (D5) and (D6).
1.2 4.2 Induction
For any integer \(0 < p \le 8 B^3\) - 1, divide \({\mathcal {H}}\) into the product
By induction on p, we will show that the closure of \(G_p\) includes the subgroup \({\hat{G}}_p\)
where \(SU(d_{p n})\) acts on the subspace \({\mathcal {Q}}_{p n}\) of \({\mathcal {Q}}_p\) with eigenvalue n of the total number operator N, and \(d_{p n}\) is the dimension of \({\mathcal {Q}}_{p n}\). The product in Eq. (D8a) is over \(0 \le n \le 2 p + 2\). Equations (D8a) and (D8b) for the case \(p = 8B^3-1\) become Eq. (D1).
The set of \(g_{zz'}\) in Eq. (D2) is a subset of the set of \(f_{xy}\) in Eq. (14) of Sect. 3. Thus \({\hat{G}}_p\) for \(p = 8B^3-1\) is a subgroup of the group G of Sect. 3. Proof of Eq. (D8a) therefore implies Eq. (18) of Sect. 3.
For \(p = 1\), Eqs. (D8a) and (D8b) follow immediately from the definition of \(K_p\). Assuming Eqs. (D8a) and (D8b) for some \(p - 1\), we will prove them for p.
Let \(S_{p n}\) be an orthonormal basis for \({\mathcal {Q}}_{p n}\) consisting of all n-fermion, m-boson, \(m \le b_{max}( p + 1)\), vectors of the form
for any list of n distinct pairs of \((z^f_i, s_i)\) and any list of m integers \(z^b_j\) such that each \(z^b_j\) coincides with at most \(b_{max} -1\) other \(z^b_{j'}\). For any pair of distinct \(|\psi _0 \rangle , |\psi _1 \rangle \in S_{p n}\), and 2 \(\times\) 2 traceless Hermitian h, define
The set of all such \(H_p( |\psi _0 \rangle , |\psi _1 \rangle , h)\) is a linear basis for the Lie algebra \(L_{p n}\) of the group \({\hat{G}}_{p n}\) of Eq. (D8b).
Thus to prove Eqs. (D8b) and (D8a) for p it is sufficient to show that any \(H_p( |\psi _0 \rangle , |\psi _1 \rangle , h)\) for some \(|\psi _0 \rangle , |\psi _1 \rangle \in S_{p n}\) and 2 \(\times\) 2 traceless Hermitian h, given the induction hypothesis, is contained in the Lie algebra generated by \(L_{p-1 n'}\) for some \(n'\) and \({\mathcal {F}}_{p-1 p}\).
1.3 4.3 Without Bosons
We consider first \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) both with m of 0 in Eqs. (D9a)–(D9c). We will work backwards starting from some \(H_p( |\psi _0 \rangle , |\psi _1 \rangle , h)\) for \(|\psi _0 \rangle , |\psi _1 \rangle \in S_{p n}\). Since \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) have the same value of total N on the region \(z \le p\), it follows that a \(U_0\) can be found in \({\hat{G}}_{p-1}\) such that
are orthogonal vectors in \(S_{p n}\), their restrictions to the region \(p-1 \le z \le p\) are also orthogonal but have equal total particle counts on \(p-1 \le z \le p\). The particle count difference between \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) at point p is at most 2, and equal and opposite to the difference between the corresponding totals on the region \(z \le p-1\). This compensating difference can be moved by \(U_0\) to the point \(p -1\).
A k in \({\mathcal {F}}_{p-1 p}\) can then be found such that
for some \(|\upsilon \rangle \in {\mathcal {H}}_p\), with particle number \(n_{\upsilon }\) and \(|\psi _6 \rangle\) and \(|\psi _7 \rangle\) orthogonal vectors in \(S_{(p-1) m}\) with \(m = n - n_{\upsilon }\).
It is then possible to find a \(U_2\) in \({\hat{G}}_{p-1}\) such that
for a some \(|\chi \rangle\) in \(S_{(p-2) (m-1)}\).
Combining Eqs. (D11a)–(D13f), the induction hypothesis implies the existence of \(U_0, U_2 \in {\hat{G}}_{p-1}\) and \(k \in {\mathcal {F}}_{(p-1) p}\) such that
The expression on the right-hand side of Eq. (D14) can then be obtained from a commutator between an operator \(k \in {\mathcal {F}}_{(p-1) p}\) and an operator \(g \in L_{(p-1) m}\) for \(m = n - n_{\upsilon }\). For 2 \(\times\) 2 traceless Hermitian \(k_{ij}\), define
and for a 2 \(\times\) 2 traceless Hermitian \(g_{ij}\), define
For any traceless, Hermitian 2 \(\times\) 2 \(h_{ij}\), there are \(k_{ij}\) and \(g_{ij}\) such that
Combining Eqs. (D14), (D15), (D16) and (D17) then gives
which completes the induction step and for \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) with m of 0 in Eqs. (D9a) - (D9c).
1.4 4.4 With Bosons
We consider next \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) both with nonzero n and m in Eqs. (D9a)–(D9c).
Suppose \(0< n < 2 p + 2\).
If the boson factors in \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) are identical, then by a combination of a rotation by a \(U_0\) in \({\hat{G}}_{p-1}\) and by a \(U_1\) in the group generated by \(k \in {\mathcal {F}}_{(p-1) p}\) the boson factors can both be turned into the case m of 0, already covered in Appendix 4.3.
Suppose the boson factors in \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) are not identical but the fermion factors in \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) are identical. Then again, but a combination of a rotation by a \(U_0\) in \({\hat{G}}_{p-1}\) and by a \(U_1\) in the group generated by \(k \in {\mathcal {F}}_{(p-1) p}\) the boson factors can both be turned into the case m of 0 but with orthogonal fermion factors in \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\). Thus back to the case covered in Appendix 4.3.
Suppose both the fermion factors and the boson factors in \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) are not identical. The induction step of Appendix 4.3 shows that the action of \({\hat{G}}_p\) is available at least on the fermion factors in \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\). A \(U_0\) in \({\hat{G}}_p\) can therefore be found which makes the fermion factors in \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) distinct both on the region \(p-1 \le z \le p\) and on the region \(0 \le z \le p - 1\). It follows that a \(U_1\) in \({\hat{G}}_{p-1}\) and a \(U_2\) in the group generated by \(k \in {\mathcal {F}}_{(p-1) p}\) can then be found which take \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) back to m of 0.
Suppose finally either n is 0 and \(|\psi _0 \rangle\) and \(|\psi _1 \rangle\) have only fermions or n is \(2 p + 2\) and all sites are filled with fermions. In either case, \({\hat{G}}_{p-1}\) and \({\hat{G}}_p\) act purely on boson states. The induction step to show that the Lie algebra of \({\hat{G}}_p\) is generated by the Lie algebra of \(L_{p-1 n'}\), either for \(n'\) of 0 or \(n'\) of 2p, and \({\mathcal {F}}_{p-1 p}\) becomes nearly a direct translation of the induction step in Appendix 4.3 from fermion states to boson states. We omit the details.
Appendix 5: Auxiliary Field Algebra
We will construct a Hilbert space \({\mathcal {H}}^B\) generated by the algebra B of polynomials in the \(\Sigma _i( x, s)\) and \(\Upsilon _i( x)\) acting purely as creation operators on \(|\Omega ^B \rangle\) and satisfying Eqs. (153a)–(153d).
Let \(B^\Sigma\) be the algebra generated by the set of all \(\Sigma _i( x, s)\), for any x, s and i, and let \(B^\Upsilon\) be the algebra generated by the set of all \(\Upsilon _i(x)\), for any x and i. Since every \(a \in B^\Sigma\) commutes with every \(b \in B^\Upsilon\), the algebra B is the tensor product
For every x, let \(B^\Sigma _x\) be the algebra generated by the set of \(\Sigma _i( x, s)\), for any s and i, and let \(B^\Upsilon _x\) be the algebra generated by the set of \(\Upsilon _i( x)\) for any i. Then for every \(x \ne y\), every \(a_x \in B^\Sigma _x\) commutes or anticommutes with every \(a_y \in B^\Sigma _y\), and every \(a_x \in B^\Upsilon _x\) commutes with every \(a_y \in B^\Upsilon _y\). Therefore the algebras \(B^\Sigma\) and \(B^\Upsilon\) are the products
Now let \(\eta _x\) be a boost that takes the point x to the point \((\tau , 0, 0, 0)\). For Eqs. (155a) and (155b) to be covariant, \(\Sigma _0(x,s)\) has to transform under boosts like \(\Psi ( x, s)\) and \(\Sigma _1(x,s)\) has to transform under boosts like \(\Psi ^\dagger ( x, s)\). Let \(S^x_{ss'}\) and \({\bar{S}}^x_{ss'}\) be the spin transformation matrices corresponding to \(\eta _x\) and define \({\hat{\Sigma }}_0(x,s)\) and \({\hat{\Sigma }}_1(x,s)\) to be
For each x and s, let \(B^\Sigma _{xs}\) be the algebra generated by \({\hat{\Sigma }}_0(x,s)\) and \({\hat{\Sigma }}_1(x,s)\). Then for \(s \ne s'\), every \(a_{xs} \in B^\Sigma _{xs}\) either commutes or anticommutes with every \(a_{xs'} \in B^\Sigma _{xs'}\). Therefore the algebra \(B^\Sigma _x\) is the product
Equation (E1) implies \({\mathcal {H}}^B\) is a tensor product
of a space generated by \(B^\Sigma\) acting on \(|\Omega ^B \rangle\) and a space generated by \(B^\Upsilon\) acting on \(|\Omega ^B \rangle\) and Eqs. (E2a) and (E2b) imply \({\mathcal {H}}^\Sigma\) and \({\mathcal {H}}^\Upsilon\) are themselves products of spaces \({\mathcal {H}}^\Sigma _x\) and \({\mathcal {H}}^\Upsilon _y\) generated, respectively, by \(B^\Sigma _x\) and \(B^\Upsilon _x\) acting on \(|\Omega ^B \rangle\)
Similarly, Eq. (E4) implies \({\mathcal {H}}^\Sigma _x\) is a product of \({\mathcal {H}}^\Sigma _{xs}\) generated by \(B^\Sigma _{xs}\) acting on \(|\Omega ^B \rangle\)
For the pair of operators \({\hat{\Sigma }}_0(x,s)\) and \({\hat{\Sigma }}_1(x,s)\) which generate \(B^\Sigma _{xs}\), Eqs. (153a) and (153c) become
Equations (E8a)–(E8c) combined with approximate Lorentz and charge conjugation invariance of the complexity of states in \({\mathcal {H}}\) imply that for the field polynomials \(P_i[ {\hat{\Sigma }}_0( x, s), {\hat{\Sigma }}_1( x, s)]\)
where u and v are normalization constants independant of x and s, an orthonormal basis for \({\mathcal {H}}^\Sigma _{xs}\) must have the form
up to an overall unitary rotation of the basis. Equations (E8a)–(E8c) imply the result of any other polynomial in \({\hat{\Sigma }}_0( x, s)\) and \({\hat{\Sigma }}_1( x, s)\) acting on \(|\Omega ^B \rangle\) is equal to some corresponding linear combination of the \(|x, s, i \rangle\) of Eq. (E10). The complexity of a state in \({\mathcal {H}}^B\) is independent of overall normalization, however, so u can be arbitrarily set to 1. The remaining constant v determines the contribution to complexity arising from sites occupied by more than a single fermion. In the continuum limit of complexity, if a continuum limit exists, the weight of multiply occupied sites in any state will go to 0. The continuum limit should therefore be independent of v.
For the pair of operators \(\Upsilon _0(x)\) and \(\Upsilon _1(x)\) which generate \(B^\Upsilon _x\), Eq. (153d) becomes
Equation (E11) combined with approximate Lorentz and charge conjugation invariance of the complexity of states in \({\mathcal {H}}\) imply that, up to an overall unitary rotation of the basis, an orthonormal basis for \({\mathcal {H}}^\Upsilon _x\) will consist of a family of states \(\{|x, n_0, n_1 \rangle ^\Upsilon \}\) labeled by a pair of nonnegative integers \(n_0, n_1\). For each \(n_0, n_1\) pair \(P_{n_0 n_1}[\Upsilon _0(x),\Upsilon _1(x)]\) is an ordered product, independent of x, of \(n_0\) copies of \(\Upsilon _0(x)\) and \(n_1\) copies of \(\Upsilon _1(x)\) subject to the requirement
The \(\{|x, n_0, n_1 \rangle ^\Upsilon \}\) are given by
where the \(u_{n_0 n_1}\) are normalization constants independant of x and symmetric in the indices \(n_0, n_1\). Equation (E11) implies the result of any other polynomial in \(\Upsilon _0( x)\) and \(\Upsilon _1( x)\) acting on \(|\Omega ^B \rangle\) is equal to some corresponding linear combination of the \(|x, n_0, n_1 \rangle\) of Eq. (E13). To be consistent with the normalization choice for fermions, \(u_{0 0}\), \(u_{0 1}\) and \(u_{1 0}\) will be set to 1. The remaining \(u_{n_0 n_1}\) determine the contribution to complexity arising from sites occupied by more than a single boson and should have no effect on the continuum limit of complexity, if a continuum limit exists.
Equation (154) implies the \(P_{n_0 n_1}[\Upsilon _0(x),\Upsilon _1(x)]\) identically vanish for \(n_0 \ge n\) or \(n_1 \ge n\).
The end result of Eqs. (E5)–(E7) is an \({\mathcal {H}}^B\) generated by the algebra B acting on \(|\Omega ^B \rangle\) which is an ordered tensor product
on which, according to Eqs. (E8a)–(E13), the \(\Sigma _i( x, s), \Upsilon _i( x),\) satisfy Eqs. (153a)–(153d).
It is convenient to define at this point an orthonormal basis P for B. In particular, no linear combination of elements of P is 0 as a result of the anticommutation and commutation relations of Eqs. (153a)–(153d). Each \(p \in P\) consists of a product of a \(p^\Sigma \in P^\Sigma\) and a \(p^\Upsilon \in P^\Upsilon\), where \(P^\Sigma\) and \(P^\Upsilon\) are orthonormal bases for the fermion field algebra \(B^\Sigma\) and the boson field algebra \(B^\Upsilon\), respectively. Each \(p^\Sigma\) is defined to be a product over all distinct x and s of one of the fermion field combinations in Eqs. (E9a)–(E9d). Each \(p^\Upsilon\) is defined to be a product over all distinct x of one of the normalized boson field combinations \(u_{n_0 n_1} P_{n_0 n_1}[\Upsilon _0(x),\Upsilon _1(x)]\).
Appendix 6: Lower Bound on the Complexity of Entangled Relativistic States
The proof of Eq. (180) bounding from below the complexity of the entangled relativistic state \(|\psi ^B \rangle\) of Eq. (178b) is a version of the proof in Appendix 2 of a lower bound on the complexity of the entangled non-relativistic state of Eq. (36), but with the regular lattice of Sect. 3.1 replaced by the random lattice of Sect. 13 and with the inclusion in \({\mathcal {H}}^B\) of anti-fermion states. The proof in Appendix 2 can be adapted to the presence of anti-fermion states in \({\mathcal {H}}^B\) by treating fermion-anti-fermion pairs in \({\mathcal {H}}^B\) following the treatment of bosons in Appendix 2. To do this we convert the complexity calculation in \({\mathcal {H}}^B\) into an equivalent complexity calculation in yet another auxiliary Hilbert space.
1.1 6.1 More Auxiliary Hilbert Spaces
For a trajectory \(k^B(\nu ) \in K^B\), let \(U_{k^B}(\nu )\) be the solution to
Define \(|\omega (\nu )^B \rangle\) to be
for a product state \(|\omega ^B(0) \rangle \in {\mathcal {H}}^B\)
with j fermions, k anti-fermions, and \(\ell\) bosons, Assume that \(|\omega ^B(0) \rangle\) and \(k^B(\nu )\) have been chosen to give
for a phase factor \(\xi\). Fermion number conservation by \(k^B(\nu )\) implies \(j - k\) must equal the fermion number n of \(|\psi ^B \rangle\).
To deal with the presence of anti-fermions in \({\mathcal {H}}^B\), we will make use of yet one more auxiliary Hilbert space, \({\mathcal {H}}^C\), which consists purely of fermion states generated by all polynomials in an auxiliary field \(\Sigma ^C_1(x,s)\) acting on an auxiliary vacuum \(|\Omega ^C \rangle\). The tensor product \({\mathcal {H}}^C \otimes {\mathcal {H}}^B\) we name \({\mathcal {H}}^D\).
There is a natural map M from \({\mathcal {H}}^D\) to \({\mathcal {H}}^B\) defined by
where \(P( \Sigma _1^C)\) is a polynomial in the field \(\Sigma _1^C(x, s)\), \(P( \Sigma _1)\) is the corresponding polynomial but in the field \(\Sigma _1(x, s)\) and \(|\psi ^B \rangle\) is any state in \({\mathcal {H}}^B\). The map M takes a subspace of \({\mathcal {H}}^D\) to the null vector in \({\mathcal {H}}^B\) and thus does not have an inverse.
Corresponding to the decomposition of \({\mathcal {H}}^D\) and \({\mathcal {H}}^B\) as tensor products over all sites
the map M is given by the product
where each \(M_x\) maps \({\mathcal {H}}^D_x\) to \({\mathcal {H}}^B_x\). The maps \(M_x\) and \(M_y\) for distinct x and y commute.
Let \(K^D\) be the Hilbert space of Hermitian operators of Sect. 15 for \({\mathcal {H}}^D\) in place of \({\mathcal {H}}^B\) and with the additional requirement that \(k^D \in K^D\) separately conserve both the fermion number \(N^B\) of \({\mathcal {H}}^B\) and the fermion number \(N^C\) of \({\mathcal {H}}^C\).
We now convert \(k^B(\nu ) \in K^B, |\omega ^B(\nu ) \rangle \in {\mathcal {H}}^B\) connecting
for a phase factor \(\xi\), to the product state \(|\omega ^B(0) \rangle\) into corresponding \(k^D(\nu ) \in K^D, |\omega ^D(\nu ) \rangle \in {\mathcal {H}}^D\) connecting some \(|\omega ^D(1) \rangle\) to a product state \(|\omega ^D(0) \rangle\) along a path such that for \(0 \le \nu \le 1\)
In addition, while \(|\omega ^B(\nu ) \rangle\) is an eigenvector of \(N^B\) with eigenvalue n, \(|\omega ^D(\nu ) \rangle\) is an eigenvector of \(N^B\) with eigenvalue 0 and of \(N^C\) with eigenvalue n. Equation (F9b) implies
Thus a lower bound on \(C^D[ |\omega ^D(1) \rangle , |\omega ^D( 0) \rangle ]\) will give a lower bound on \(C^B[ |\omega ^B(1) \rangle , |\omega ^B( 0) \rangle ]\).
Let the product state \(|\omega ^D(0) \rangle\) be \(|\omega ^C \rangle \otimes |\omega ^B \rangle\) where
for \(d_f( p_i)\), \(d_{{\bar{f}}}( q_i)\) and \(d_b( r_i)\) from Eq. (F3), and \(d_f^C( p_i)\) constructed from \(d_f( p_i)\) of Eq. (F3) by substituting \(\Sigma _1^C(x, s)\) for \(\Sigma _1(x, s)\).
Equations (F8) and (F9a) imply the state \(|\omega ^D(1) \rangle\) will satisfy
In addition, since the trajectory \(k^D(\nu )\) conserves \(N^B\) and \(N^C\) and \(|\omega ^D(0) \rangle\), by Eqs. (F11a) and (F11b), has \(N^B\) of 0 and \(N^C\) of n, \(|\omega ^D(1) \rangle\) must have these same eigenvalues. Also, since M acts only on the \(\Sigma _1^C(x,s)\) fermion content of \(|\omega ^D(1) \rangle\) and \(|\psi \rangle ^B\), by Eqs. (178a) and (178b), has no boson content and no \(\Sigma _0^B(x,s)\) anti-fermion content, \(|\omega ^D(1) \rangle\) must have no boson content, no \(\Sigma _0^B(x,s)\) and \(\Sigma _1^B(x,s)\) content and be given instead by
where \(|\psi ^C \rangle\) is
for \(p^C_i\) given by
for the same \(\zeta _i, u^k(x)\) and \(D_{ij}\) in Eqs. (175)–(178b) for \(|\psi ^B \rangle\).
For both the nonrelativistic version of complexity in Sect. 3.3 and the relativistic version in Sect. 16, \(C( |\psi \rangle )\) is actually independent of the normalization of \(|\psi \rangle\). We can therefore safely set z to 1 in Eq. (F14a). The result is that \(|\omega ^D(1) \rangle\) in Eq. (F13) is normalized to 1, which for consistency we now assume also for \(|\omega ^D(0) \rangle\).
Now approximate Eqs. (17a), (17b), (B1) and (B2) for \(|\omega ^B(\nu ) \rangle\) by a series of discrete steps
We will prove Eqs. (F9a) and (F9b) by induction in \(\nu\). Eqs. (F5), (F3), (F11a) and (F11b) give Eq. (F9a) and (F9b) for \(\nu = 0\)
Now assume \(k^D( \nu )\) satisfying Eq. (F9b) has been found for \(\nu < \nu '\) such that \(|\omega ^D(\nu ) \rangle\) given by
satisfies Eq. (F9a) for \(\nu \le \nu '\). We will show that a \(k^D( \nu ')\) exists satisfying Eq. (F9b) and extending Eq. (F9a) to \(\nu ' + \delta\).
According to Eq. (A8), \(k^B\) in Eq. (F16) consists of a sum of operators of the form
where \(f^B_{xy}\) is a Hermitian operator on \({\mathcal {H}}^B_x \otimes {\mathcal {H}}^B_y\) for a pair of nearest neighbor sites \(\{x,y\}\) which conserves \(N^B\) and has vanishing partial traces for both \({\mathcal {H}}^B_x\) and \({\mathcal {H}}^B_y\). We assume the dimension \(d_{\mathcal {H}}\) of \({\mathcal {H}}^D_x\), and the corresponding slightly smaller dimension of \({\mathcal {H}}^B_x\), are large enough that the contribution to \(k^B\) of single site operators of the form given in Eq. (A7a) can be neglected.
Then the required \(k^D( \nu ')\) can be found if for every allowed \({\hat{f}}^B_{xy}\) there is a \({\hat{f}}^D_{xy}\) of the form
where \(f^D_{xy}\) is a Hermitian operator on \({\mathcal {H}}^D_x \otimes {\mathcal {H}}^D_y\) which has vanishing partial traces for both \({\mathcal {H}}^D_x\) and \({\mathcal {H}}^D_y\), conserves \(N^B\) and \(N^C\) and for which
To find the required \(f^D_{xy}\), decompose \({\mathcal {H}}^D_x \otimes {\mathcal {H}}^D_y\) into a direct sum of subspaces
with eigenvalues m and n of \(N^B\) and \(N^C\), respectively. Similarly, decompose \({\mathcal {H}}^B_x \otimes {\mathcal {H}}^B_y\) into a direct sum of subspaces
with eigenvalue m of \(N^B\).
Let \(P^D_{mn}\) be the projection operator onto \({\mathcal {H}}^D_{mn}\). Define \(M_{mn}\) to be
for \(M_x\) and \(M_y\) from Eq. (F7). Then \(M_{mn}\) maps \({\mathcal {H}}^D_{mn}\) onto \({\mathcal {H}}^B_{m+n}\). Let \({\mathcal {H}}^{D\perp }_{mn}\) be the orthogonal complement of the subspace of \({\mathcal {H}}^D_{mn}\) which is mapped to 0 by \(M_{mn}\). For each \(|\psi ^B \rangle \in {\mathcal {H}}^B_{m+n}\) there is a unique \(|\psi ^D \rangle \in {\mathcal {H}}^{D\perp }_{mn}\) such that
For each such \(|\psi ^B \rangle\), define \(M_{mn}^{-1}\)
and for any \(|\psi ^B \rangle \in {\mathcal {H}}^B_\ell\) with \(\ell\) other than \(m + n\)
where \(P^B_{m+n}\) is the projection operator onto \({\mathcal {H}}^B_{m+n}\). Define \(g^D_{xy}\) to be
By Eq. (F23), \(g^D_{xy}\) maps each \({\mathcal {H}}^D_{mn}\) into itself and therefore conserves both \(N^C\) and \(N^B\).
We then have
where the first line follows from by Eq. (F27) and the second follows because \(M_{mn}\) maps onto \({\mathcal {H}}^B_{m+n}\) and \(f^B_{xy}\) conserves \(N_B\). Equations (F7), (F29) and the induction hypothesis, Eq. (F9a) for \(\nu '\), give
which is Eq. (F20a).
In addition, since \(M_{mn}^{-1}\) maps into \({\mathcal {H}}^D_{mn}\), Eqs. (F23) and (F27) imply
We then have
where the first line follows from Eqs. (F28) and (F31) and the second holds because \(f^B_{xy}\) conserves \(N^B\). Since the index s of \(\Sigma _1(x,s)\) is in the range \(0 \le s < 4\), the maximum possible value of \(N^B\) for x and y together is 8. As a result there are at most 9 different combinations of m and n giving any value of \(m + n\). Equation (F32) then implies
which is Eq. (F20b).
Finally, \(g^D_{xy}\) can be split into
where
and \(d_{\mathcal {H}}\) is the dimension of each \({\mathcal {H}}^D_x\). Eqs. (F34)–(F35d) imply
Equations (F36) and (F33) imply \(f^D_{xy}\) satisfies Eq. (F20b). For \(d_{\mathcal {H}}\) large enough, Eqs. (F34) and (F30) imply \({\hat{f}}^D_{xy}\) satisfies Eq. (F20a).
Which completes the induction step of the proof of Eqs. (F9a) and (F9b). Equation ( F10) follows. To obtain a lower bound on \(C^B[ |\omega ^B(1) \rangle , |\omega ^B( 0) \rangle ]\) we now derive a lower bound on \(C^D[ |\omega ^D(1) \rangle , |\omega ^D( 0) \rangle ]\).
1.2 6.2 Schmidt Spectra Again
From each region \(D_{ij}\), extract a subset \({\hat{D}}_{ij}\), consisting of the center points x of all cells c(x) reached by starting at \(y_{ij}\) and traveling along a geodesic in \(L(\tau , \sigma )\) in the positive or negative \(x_1\) direction a number \(\le \frac{d}{2 \rho }\) of discrete steps each of proper length \(2 \rho\), then traveling along a geodesic in the positive or negative \(x_2\) direction a number \(\le \frac{d}{2 \rho }\) of discrete steps each of proper length \(2 \rho\), then traveling along a geodesic in the positive or negative \(x_3\) direction a number \(\le \frac{d}{2 \rho }\) of discrete steps each of proper length \(2 \rho\). Since each c(x) is contained within a sphere of radius \(\rho\) around its center point, none of the points in \({\hat{D}}_{ij}\) will be nearest neighbors and for large d, the total number of points in each \({\hat{D}}_{ij}\) will be nearly \(\frac{d^3}{\rho ^3}\). Since V is between \(\frac{48 d^3}{\pi \rho ^3}\) and \(\frac{ 6 d^3}{\pi \rho ^3}\), the number of points in each \({\hat{D}}_{ij}\) is zV for z between \(\frac{\pi }{6}\) and \(\frac{\pi }{48}\). We will assume V is large enough that we can ignore the statistical uncertainty in the number of points in each \({\hat{D}}_{ij}\).
From this set of \({\hat{D}}_{ij}\) construct a set of subsets \(E_\ell\) each consisting of 2n distinct points chosen from 2n distinct \({\hat{D}}_{ij}\). Since there are mn sets \({\hat{D}}_{ij}\), there will be \(\frac{z m V}{2}\) sets \(E_\ell\).
For each \(E_\ell\) form the tensor product spaces
The space \({\mathcal {Q}}_\ell\) has dimension \(16^{2n}\) and \({\mathcal {H}}^D\) becomes
A Schmidt decomposition of \(|\omega ^D(\nu ) \rangle\) according to Eq. (F38) then becomes
where
for \(0 \le j < 16^{2n}\) and real non-negative \(\lambda _{j\ell }( \nu )\) which fulfill the normalization condition
Each \(|\phi _{j\ell }(\nu ) \rangle\) is a pure fermion state while the \(|\chi _{j\ell }(\nu ) \rangle\) can include fermions, antifermions and bosons.
The fermion number operators \(N^C[{\mathcal {Q}}_\ell ]\) and \(N^C[{\mathcal {R}}_\ell ]\) commute and \(|\omega ^D(\nu ) \rangle\) is an eigenvector of the sum with eigenvalue n. It follows that the decomposition of Eq. (F39) can be done with \(|\phi _{j\ell }( \nu ) \rangle\) and \(|\chi _{j\ell }(\nu ) \rangle\) eigenvectors of \(N^C[{\mathcal {Q}}_\ell ]\) and \(N^C[{\mathcal {R}}_\ell ]\), respectively, with eigenvalues summing to n. Let \(|\phi _{0\ell } \rangle\) be \(|\Omega _\ell \rangle\), the vacuum state of \({\mathcal {Q}}_\ell\), and let \(|\phi _{i\ell } (\nu ) \rangle , 1 \le i \le 8n\), span the 8n-dimensional subspace of \({\mathcal {Q}}_\ell\) with \(N^C[{\mathcal {Q}}_\ell ]\) of 1. We assume the corresponding \(\lambda _{i\ell }( \nu ), 1 \le i \le 8n\), are in nonincreasing order. Consider Eqs. (F13)–(F15) for \(|\omega ^D(1) \rangle\). For any choice of \(\ell\) there will be a set of 2n nonzero orthogonal \(|\phi _{1\ell }( 1) \rangle , \ldots |\phi _{2n\ell }( 1) \rangle\) with
On the other hand, for the product state \(|\omega ^D(0) \rangle\) in Eqs. (F11a) and (F11b), the \(|\phi _{j\ell }(\nu ) \rangle\) come entirely from \(|\omega ^C \rangle\), which is a product of n independent single fermion states. The space spanned by the projection of these into some \({\mathcal {Q}}_\ell\) is at most n dimensional, and as a result at most n orthogonal \(|\phi _{1\ell }(0) \rangle ,\ldots |\phi _{n\ell }(0) \rangle\) can occur. Therefore at \(\nu = 0\), there will be at most n nonzero \(\lambda _{1\ell }(0), \ldots \lambda _{n\ell }(0)\). For \(n < j \le 8n\), we have
Since \(\{\lambda _{j\ell }( \nu )\}\) is a unit vector, Eqs. (F42) and (F43) imply that as \(\nu\) goes from 0 to 1, \(\{\lambda _{j\ell }( \nu )\}\) must rotate through a total angle of at least \(\arcsin (\sqrt{\frac{n}{mV}})\).
For the small interval from \(\nu\) to \(\nu + \delta \nu\) let \(\mu _{j\ell }(\nu )\) and \(\theta _{\ell }(\nu )\) be
We then have
Summed over the \(\frac{zmV}{2}\) values of \(\ell\), Eq. (F45) becomes
and therefore
since z is greater than \(\frac{\pi }{48}\).
1.3 6.3 Rotation Matrix and Rotation Angle Bounds
The rest of the proof of the lower bound on relativistic complexity, Eq. (180), is a copy of Sections 2.3 and 2.4 of the proof in Appendix 2 of the non-relativistic lower bound, Eq. (37), but with the non-relativistic fermion charge N and Hilbert spaces \({\mathcal {H}}^f\) and \({\mathcal {H}}^b\) replaced, respectively, by \(N^C\), \({\mathcal {H}}^C\) and \({\mathcal {H}}^B\).
As in Appendix 2.3, the rotation of \(\lambda _{j \ell }(\nu )\) during the interval from \(\nu\) to \(\nu + \delta \nu\) will be determined by the sum \(g^D_{\ell }(\nu )\) of all contributions to \(k^D(\nu )\) of Eq. (F17) arising from \(f^D_{xy}\) for nearest neighbor pairs \(\{x,y\}\) with one point, say x, in \(E_\ell\). By construction of the \(E_\ell\), if x is in \(E_\ell\), y can not be in \(E_\ell\) or any distinct \(E_{\ell '}\). A repeat of the derivation of Eqs. (B36)–(B38) then leads to
for the rotation matrix \(r_{jk\ell }(\nu )\)
for \(|\phi _{k\ell }(\nu ) \rangle\) and \(|\chi _{k\ell }(\nu ) \rangle\) of Eq. (F39) and \(\mu _{j\ell }(\nu )\) of Eq. (F44a).
Since the \(f^D_{xy}\) contributing to \(g^D_\ell (\nu )\) conserve total fermion number \(N^C\), \(g^D_\ell (\nu )\) can be expanded as
where \(z^0( x, y, \nu )\) acts only on states with \(N^C( {\mathcal {H}}^D_x \otimes {\mathcal {H}}^D_y)\) of 0, \(z^1( x, y, \nu )\) acts only on states with \(N^C( {\mathcal {H}}^D_x \otimes {\mathcal {H}}^D_y)\) strictly greater than 0, and the \(z^i(x,y,\nu )\) are normalized by
The operator \(z^0(x, y, \nu )\) will be
where \(P^C(x,y)\) projects onto the vacuum state of \({\mathcal {H}}^C_x \otimes {\mathcal {H}}^C_y\) and \(g^B(x,y,\nu )\) is a normalized Hermitian operator acting on \({\mathcal {H}}^B_x \otimes {\mathcal {H}}^B_y\)
Combining Eqs. (F44b),(F48) - (F50b) gives
with the definitions
A duplicate of the proof of Eqs. (B46)–(B52) then yields
which combined with Eq. (F53a) imply
where
The Cauchy–Schwartz inequality then gives
A repeat of the argument leading to Eq. (B56) implies
where M is the maximum number of nearest neighbors of any lattice point x. An upper bound on M can be found as follows. Recall any c(x) is contained in a sphere with center x and radius \(\rho\) and contains a sphere with center x and radius \(\frac{\rho }{2}\). It follows that M is less than or equal to the number of disjoint spheres of radius \(\frac{\rho }{2}\) that can placed with centers on a sphere with center x and radius \(2 \rho\). For each of the \(\frac{\rho }{2}\) spheres, a slice through its center orthogonal to the line from its center to x will be contained in a sphere with center x and radius \(\frac{\sqrt{17}}{2}\). The area of each of these slices is \(\frac{\pi \rho ^2}{4}\), the area of the radius \(\frac{\sqrt{17}}{2}\) sphere is \(\frac{68 \pi \rho ^2}{4}\), and therefore
By Eq. (16)
In addition, \(z^0(x,y,\nu )\) is orthogonal to \(z^1(x, y, \nu )\). It follows that
Assembling Eqs. (F58), (F59), (F60) and (F62) gives
Eq. (F47b) then implies
and therefore
which by Eqs. (F10) and (F8) yields
Since Eq. (F66) holds for all product \(|\omega ^B(0) \rangle\) we finally obtain
Appendix 7: Upper Bound on the Complexity of Entangled Relativistic States
The proof of Eq. (181) bounding from above the complexity of the entangled relativistic state \(|\psi ^B \rangle\) of Eq. (178b) follows the proof in Appendix 3 of an upper bound on the complexity of the entangled non-relativistic state of Eq. (36), but with the regular lattice of Sect. 3.1 replaced by the random lattice of Sect. 13 and \({\mathcal {H}}\) replaced by \({\mathcal {H}}^B\).
An upper bound on \(C^B( |\psi ^B \rangle )\) is given by \(C^B( |\psi ^B \rangle , |\omega ^B \rangle )\) for any product state \(|\omega ^B \rangle\), for which in turn an upper bound is given by
for any trajectory \(k^B(\nu ) \in K^B\) fulfilling
for a phase factor \(\xi\).
As in Appendix 3, construction of a sufficient \(k^B(\nu )\) begins with an \(|\omega ^B \rangle\) consisting of n fermions each at one of a corresponding set of n single points. Then \(|\omega ^B \rangle\) is split into a sum of m orthogonal product states, each again consisting of n fermions one at each of a corresponding set of n single points. Then the position of each of the fermions in the product states is moved to the center of of one of the monomials of Eq. (177). Finally, by approximately \(\ln ( V) / \ln ( 8)\) iterations of a fan-out tree, the mn wave functions concentrated at points are spread over the mn cubes \(D_{ij}\).
1.1 7.1 Cell Count Bound
The bound on \(C^B( |\psi ^B \rangle )\) relies on a bound we will derive first on the number of distinct cells \(c(x), x \in L( \tau , \sigma , \rho )\), which intersect a geodesic \(g( \lambda ) \in L( \tau , \sigma ), 0 \le \lambda \le \lambda _{max},\) of length \(\lambda _{max}\).
Let \({\bar{g}}\) be the set of all points within a proper distance \(2 \rho\) of any point on \(g(\lambda )\). Since every c(x) is contained in a sphere with center x and radius \(\rho\), it follows that \({\bar{g}}\) contains all c(x) which intersect \(g(\lambda )\). On the other hand, each c(x) within \({\bar{g}}\) contains a sphere with center x and radius \(\frac{\rho }{2}\) which is disjoint from all other \(c(x')\) contained in \({\bar{g}}\). The total volume occupied by the collection of disjoint radius \(\frac{\rho }{2}\) spheres has to be less than the total volume of \({\bar{g}}\). The number \(p(\lambda _{max})\) of c(x) which intersect \(g(\lambda )\) is therefore bounded by
1.2 7.2 Product State to Entangled State
For each value of \(0 \le i < m\), let \(x_{i0}\) be the center point of the cell found by traveling from an arbitrarily chosen starting point, \(x_{00}\), along a geodesic in the \(x^1\) direction a proper distance of \(4 i \rho\). Then from each \(x_{i0}\) travel along a geodesic in the \(x^2\) direction. For each \(0< j < n\), let \(x_{ij}\) be the center point of the cell the geodesic beginning at \(x_{i0}\) enters after leaving the cell with center point \(x_{ij-1}\). All points on the geodesics beginning at \(x_{i0}\) and at \(x_{i'0}\) for \(i \ne i'\) will be at least a distance of \(4 \rho\) apart. As a result each \(x_{ij}\) will be both distinct from and not a nearest neighbor of each \(x_{i'j'}\) with \(i \ne i'\). The gap between \(x_{ij}\) and \(x_{i'j'}\) accomplishes the goal of making it possible, despite the random placement of cells, to insure that \(x^B_{ij}\) and \(x^B_{ij+1}\) are nearest neighbors as will turn out to be required.
Let the set of n-particle product states \(|\omega ^B_i \rangle\) be
The entangle n-particle state \(|\chi ^B \rangle\)
we generate from a sequence of unitary transforms acting on \(|\omega ^B \rangle = |\omega ^B_0 \rangle\).
The sequence of \(k^B\) which convert the product state \(|\omega ^B \rangle\) into the entangled state \(|\chi ^B \rangle\) follows the sequence of k mapping the product state \(|\omega \rangle\) to the entangled state \(|\chi \rangle\) in Appendix 3.1, with the non-relativistic fermion operator \(\Psi ^\dagger ( x, s)\) replaced by the relativistic \({\hat{\Sigma }}_1( x, s)\).
From \(k^B_0, \ldots k^B_{n-2}\) in place of \(k_0, \ldots k_{n-2}\) of Eqs. (C5)–(C13) we obtain
with
as in Eqs. (C18a) and (C18b) and therefore total cost
The spinor \(v^k(x)\) in Eq. (G6) , as defined in Section 17, is orthogonal to \(u^k(x)\) of Eq. (G4) and obtained, as is \(u^k(x)\), by boosting from the origin of \(L( \tau , \sigma )\) to x a spin state of a free fermion at rest at the origin of \(L( \tau , \sigma )\).
Then from \(k^B_{n-1}, \ldots k^B_{n - 1 +p}, 3n - 2 \ge p < 48 n^2 + 159 n\), in place of \(k_{n-1}, \ldots k_{2n-2}\) of Eqs. (C14a)–(C15) we obtain
with \(\parallel k^B_i \parallel , |\theta _i|\) satisfying Eqs. (G7a) and (G7b) and incremental cost
The count of additional \(k^B_i\) required for Eq. (G9) arises as follows. A geodesic between \(x^B_{ij}\) and \(x^B_{i+1j}\) has proper length \(\lambda\) in the range \(2 \rho \le \lambda < (2 n + 4) \rho\) and therefore, according to Eq. (G3), can pass through a total of between 3 and \(48 n + 160\) cells, and thus requires between 2 and \(48 n + 159\) nearest neighbor steps. The sequence of \(k^B_{n-1}, \ldots k^B_{n - 1 +p}\) for the map of Eq. (G9) can be required to complete between 2 and \(48 n + 159\) such steps from \(x^B_{ij}\) and \(x^B_{i+1j}\) for each \(0 \ge j < n\), hence \(3n - 2 \ge p < 48 n^2 + 159 n\).
Following Eqs. (C16) and (C17), we now apply copies of the maps of Eqs. (G6) and (G9) along the \(x^2\) direction geodesics at \(x_{10}, \ldots x_{m0}\) with end result
where all \(k^B_i\) satisfy Eq. (G7a), \(\theta ^B_i\) satisfy Eq. (G7b) and
The cost of the transition from \(|\omega ^B \rangle\) to \(|\chi ^B \rangle\) is then bounded by
1.3 7.3 Entangled State Repositioned
Let the entangled n-particle state \(|\phi ^B \rangle\) be
where, as defined in Section 17, \(y_{ij}\) is the center of cube \(D_{ij}\) in Eq. (175) and \(\zeta _i\) is the phase factor of monomial \(p_i\) in Eq. (176a).
Equations (C21a)–(C32) translate directly from the non-relativistic field theory to the relativistic case, with the result
The distance r is given by
where \(r_{ij}\) is the number of nearest neighbor steps in the shortest path between lattice points \(x_{ij}\) and \(y_{ij}\) such that no pair of paths for distinct \(\{i, j\}\) intersect, \(y_{ij}\) is the center point of \(D_{ij}\) and \(x_{ij}\) is the \(m \times n\) grid of points of Appendix 7.2.
1.4 7.4 Fan-Out
Following Appendix 3.3 of the proof of the non-relativistic upper bound in Appendix 3, the state \(|\phi ^B \rangle\) with particles at the centers of the cubes \(D_{ij}\) we now fan-out to the state \(|\psi ^B \rangle\) with particle wave functions spread uniformly over the cubes \(D_{ij}\). For sufficiently small \(\rho\) nearly all of the complexity in the bound on \(C^B(|\psi ^B \rangle )\) is generated in this step.
We will construct a fan-out initially for \(D_{00}\), which will then be duplicated on the remaining \(D_{ij}\). Recall the \(x \in D_{00}\) are the centers of all cells crossed by starting at \(y_{00}\) and traveling along a geodesic in the positive or negative \(x^1\) direction a proper distance of less than d, then in the positive or negative \(x^2\) direction a proper distance less than d, then in the positive or negative \(x^3\) direction a proper distance less than d.
The set of \(x \in D_{00}\) we will arrange as the endpoints of a tree constructed in a sequence of stages most of which increase the number of endpoints of the tree by a factor of 8. Starting at \(y_{00}\), travel along a geodesic in the positive or negative \(x^1\) directions a proper distance of \(\frac{d}{2}\). Define this set of 2 points to be s(1). From each of the points of s(1) , travel along a geodesic in the positive or negative \(x^2\) direction a proper distance of \(\frac{d}{2}\). Let this set of 4 points be s(2). From each of the points of s(2) , travel along a geodesic in the positive or negative \(x^3\) direction a proper distance of \(\frac{d}{2}\). The resulting set of 8 points is s(3). Repeating this sequence of 3 steps a total of p times yields a set s(3p) of \(8^p\) endpoints, each a distance of \(\frac{d}{2^{p-1}}\) from its nearest neighbor. For each \(y \in s(3p)\) let \({\hat{s}}(y)\) be the set of 8 \(y' \in s(3p +3)\) reached by a sequence of geodesic segments originating at y.
Now choose q such that
Each pair of distinct points in s(3q) will be separated by a distance of at least \(2 \rho\). Since every cell c(x) is contained in a sphere of radius \(\rho\) around x, each \(y \in s(3q)\) will lie in a distinct cell. Similarly, for all \(r < q\), each \(y \in s(3r)\) will lie in a distinct cell. For each \(y \in s(3r), r \le q\), let x(y) be the center point of the cell containing y
At the outset of Sect. 17 we assumed \(\rho\) is much smaller than the proper time \(\tau\) of the hyperboloid \(L( \tau , \sigma )\). The region in \(L( \tau , \sigma )\) occupied by a collection of nearby \(y \in s(3q)\) will therefore be nearly flat and can be divided up into disjoint cubes each with edge length \(\frac{d}{2^{q-1}}\) centered on a corresponding \(y \in s(3q)\). Let the cube for \(y \in s(3q)\) be t(y). The union of all t(y) covers \(D_{00}\). Let w(y) be
Define n(y) to be the number of points in w(y). Working backwards iteratively from s(3q), define n(y) for \(y \in s(3p), p < q\), by
Carried back to \(n(y_{00})\) the result is V, the total number of points in \(D_{00}\).
For any \(r \le q\), define the state \(\upsilon ^B_{3r}\) to be
Equations (C35a)–(C49) of the non-relativistic fan-out process in Appendix 3.3 can then be adapted to generate a sequence of \(\exp ( i \frac{\pi }{2} k^B)\) which map \(|\upsilon ^B_{3r-3} \rangle\) into \(|\upsilon ^B_{3r} \rangle\). For the non-relativistic fan-out process, each step in which a state is split yields a pair of equally weighted pieces. For the splitting process corresponding to the states of Eq. (G20) the resulting pair will not in general be weighted equally, but this difference by itself does not affect the complexity bound. In the course of the map taking \(|\upsilon ^B_{3r-3} \rangle\) into \(|\upsilon ^B_{3r} \rangle\) , each of the 3 geodesic segments by which any point in s(3r) is reached from its parent point in \(s(3r - 3)\) will be of length \(\frac{d}{2^r}\). Equation (G3) implies that the number of nearest neighbor steps to traverse a segments of length \(\frac{d}{2^r}\) is bounded by \(24 \frac{d}{\rho 2^r} + 63\). A repetition of the derivation of Eq. (C50) then yields
The last step in the fan-out process consists of splitting the piece of \(|\upsilon ^B_{3q} \rangle\) at each x(y) into n(y) equally weighted components, then distributing these across the cube t(y) to produce the state
The complexity of the map taking \(|\upsilon ^B_{3q} \rangle\) to \(|\upsilon ^B_{3q+1} \rangle\) can be bounded as follows. For each \(y \in s(3q)\) the length of the shortest line connecting the cell holding y to the cell holding any \(x \in w(y)\) is bounded by \(\frac{\sqrt{3} d}{2^q}\), the distance from y to a corner of t(y), which according to Eq. (G17) is bounded by \(2 \sqrt{3} \rho\). Equation (G3) implies that the number of nearest neighbor steps to traverse a segment of length \(2 \sqrt{3} \rho\) is bounded by \(48 \sqrt{3} + 63\). For any \(x \in w(y)\), at each \(z \in w(y)\) along the path from x to y, the remaining path from z to y is the shortest nearest neighbor route to y. It follows that if the paths from some \(x \in w(y)\) to y and from a distinct \(x' \in w(y)\) to y coincide at z the remaining segments from z to y will also coincide. The collection of shortest paths from all \(x \in w(y)\) to y must therefore form a tree, each branch of which consists of at most \(48 \sqrt{3} + 63\) nearest neighbor steps. By adapting the derivation of Eq. (C50) the cost of all such paths executed in parallel for all \(x \in D_{00}\), the total count of which is V, can then be bounded to give
Summing Eq. (G21) over r from 1 to q, adding Eq. (G23) and using Eq. (G17) gives
where
The bound of Eq. (G24) applies to a fan-out process on a single cube \(D_{00}\). Assume the process repeated in parallel on the mn cubes \(D_{ij}\), thereby generating \(|\psi ^B \rangle\) of Eq. (178b). For \(|\phi ^B \rangle\) of Eq. (G14) we then have
From Eqs. (G13) and (G15), it follows that for a product state \(|\omega ^B \rangle\) we have
for \(c_1\) of Eq. (G25), r of Eq. (G16) and
Equation (181) then follows.
Rights and permissions
About this article
Cite this article
Weingarten, D. Macroscopic Reality from Quantum Complexity. Found Phys 52, 45 (2022). https://doi.org/10.1007/s10701-022-00554-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10701-022-00554-0