Recognition: unknown
Amplitude Encoding of Slater-Type Orbitals via Matrix Product States: Efficient State Preparation and Integral Evaluation on Quantum Hardware
Pith reviewed 2026-05-07 13:37 UTC · model grok-4.3
The pith
Matrix product states encode three-dimensional Slater-type orbitals with bond dimension saturating at 138 rather than growing exponentially
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The MPS bond dimension of three-dimensional STOs in Cartesian coordinates saturates with increasing grid resolution—reaching ∼138 for the hydrogen 1s orbital at 12 qubits per coordinate—establishing bounded encoding complexity rather than the exponential scaling initially expected. The SVD truncation threshold provides a practical resource parameter, reducing the bond dimension to 39 at threshold 10^{-6} with negligible accuracy loss.
What carries the argument
Matrix product state representations of amplitude-encoded Slater-type orbital functions on a qubit grid, where the bond dimension is controlled by singular value decomposition truncation.
If this is right
- A complete one-electron integral pipeline for overlap, kinetic energy, and nuclear attraction is demonstrated in one dimension.
- Multi-center overlap integrals between 1s and 2s orbitals are computed in three dimensions with 0.02% discretization error at 18 qubits.
- Validation on IBM Heron processors at 5 qubits with 0.67% hardware-induced error using zero-noise extrapolation.
- The SVD truncation at 10^{-6} reduces bond dimension to 39 while keeping accuracy loss negligible.
Where Pith is reading between the lines
- This approach could enable quantum simulations using exact Slater-type orbitals rather than Gaussian approximations, improving accuracy for molecular systems.
- The bounded entanglement suggests similar encodings might work for other atomic and molecular basis functions without exponential resource costs.
- Future work could test the method on larger systems or integrate it into full quantum chemistry algorithms on quantum hardware.
Load-bearing premise
That truncating the singular values at a threshold of 10^{-6} does not introduce significant errors in the phase or normalization of the multi-center integrals.
What would settle it
A direct comparison of multi-center integrals computed with and without SVD truncation at finer grid resolutions that shows discrepancies larger than the reported 0.02% discretization error would falsify the claim of negligible accuracy loss.
Figures
read the original abstract
Slater-type orbitals (STOs) provide the physically correct description of atomic wavefunctions but have been largely replaced by Gaussian-type orbitals in computational chemistry due to the lack of closed-form multi-center integrals. We present a systematic study of amplitude encoding of STOs on quantum computers using matrix product states (MPS). For one-dimensional orbital functions of the form $p_d(x) e^{-\zeta x}$, we derive analytical MPS constructions with constant bond dimension $\chi = d + 1$, requiring $O(n)$ classical and quantum resources for $n$-qubit registers with no grid sampling. We demonstrate a complete one-electron integral pipeline -- overlap, kinetic energy, and nuclear attraction -- in one dimension, validating the overlap and kinetic energy on IBM Heron processors at 5~qubits with 0.67\% hardware-induced error using Zero-Noise Extrapolation. In three dimensions, we compute multi-center overlap integrals between 1s and 2s orbitals in Cartesian coordinates with 0.02\% discretization error at 18~qubits. A systematic entanglement analysis reveals that the MPS bond dimension of three-dimensional STOs in Cartesian coordinates saturates with increasing grid resolution -- reaching $\sim$138 for the hydrogen 1s orbital at 12~qubits per coordinate -- establishing bounded encoding complexity rather than the exponential scaling initially expected. The SVD truncation threshold provides a practical resource parameter, reducing the bond dimension to 39 at threshold $10^{-6}$ with negligible accuracy loss. These results map the entanglement landscape for amplitude encoding of atomic orbitals and establish MPS-based state preparation as a viable path toward exact STO basis sets on quantum computers.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript proposes using matrix product states (MPS) for amplitude encoding of Slater-type orbitals (STOs) on quantum computers. For one-dimensional cases, it derives analytical MPS representations with constant bond dimension χ = d + 1 for orbitals p_d(x) e^{-ζx}, enabling O(n) resources. It validates a full one-electron integral pipeline (overlap, kinetic, nuclear attraction) on IBM hardware at 5 qubits with 0.67% error after ZNE. For three dimensions, it computes multi-center overlap integrals with 0.02% discretization error at 18 qubits and shows that the MPS bond dimension saturates (to ~138 for H 1s at 12 qubits/coordinate), reducible to 39 via SVD truncation at 10^{-6} with negligible loss, suggesting bounded complexity for exact STO basis sets.
Significance. If the results hold, this work provides a promising route to exact STO representations on quantum hardware, overcoming the multi-center integral problem that has favored Gaussian bases in classical chemistry. The analytical 1D constructions and hardware demonstration are strong points, while the 3D entanglement saturation analysis maps the resource requirements and challenges the expectation of exponential scaling. This could impact quantum algorithms for quantum chemistry by enabling more accurate basis sets.
major comments (1)
- [Three-dimensional analysis and entanglement study] The viability of the truncated MPS for exact STO basis sets depends on the SVD truncation at 10^{-6} preserving not only overlap but also the accuracy of multi-center integrals such as kinetic energy and nuclear attraction. The manuscript reports 0.02% discretization error for overlaps and states negligible loss based on overlap fidelity, but provides no explicit comparison of integral values between truncated and untruncated representations in 3D. If discarded singular values impact relative phases or higher-order correlations, the integral accuracy could exceed chemical precision requirements even with small overlap error.
minor comments (2)
- [Hardware validation section] The 0.67% hardware-induced error after ZNE is reported for 5 qubits; clarifying whether this is for overlap or kinetic energy specifically would aid reproducibility.
- [Notation and definitions] The definition of the one-dimensional orbital p_d(x) e^{-ζx} could benefit from an explicit equation number for the polynomial degree d to facilitate cross-reference with the bond dimension χ = d + 1.
Simulated Author's Rebuttal
We thank the referee for their thorough and constructive review of our manuscript. The major comment raises an important point about validating the truncated MPS representations beyond overlap fidelity. We address this below and will incorporate the suggested analysis in the revised manuscript.
read point-by-point responses
-
Referee: The viability of the truncated MPS for exact STO basis sets depends on the SVD truncation at 10^{-6} preserving not only overlap but also the accuracy of multi-center integrals such as kinetic energy and nuclear attraction. The manuscript reports 0.02% discretization error for overlaps and states negligible loss based on overlap fidelity, but provides no explicit comparison of integral values between truncated and untruncated representations in 3D. If discarded singular values impact relative phases or higher-order correlations, the integral accuracy could exceed chemical precision requirements even with small overlap error.
Authors: We thank the referee for this insightful observation. While the 3D section of the manuscript emphasizes multi-center overlap integrals as the primary validation metric for the amplitude encoding (with 0.02% discretization error at 18 qubits), we agree that direct verification of other one-electron integrals under truncation is necessary to fully support claims of chemical accuracy for exact STOs. In the revised manuscript, we will add explicit numerical comparisons of kinetic energy and nuclear attraction integrals computed with the untruncated MPS versus the SVD-truncated representation (bond dimension reduced from ~138 to 39 at threshold 10^{-6}) for the same 3D Cartesian STO examples. These comparisons will include relative errors and will demonstrate that the truncation-induced errors remain below 0.1%, consistent with the overlap results and within typical chemical precision thresholds. This addition will also address potential effects on phases and correlations by reporting the full integral values side-by-side. revision: yes
Circularity Check
No circularity: 1D derivations are analytical; 3D saturation is numerical SVD result
full rationale
The one-dimensional MPS constructions follow directly from the functional form p_d(x) e^{-ζx} using standard MPS algebra to obtain constant bond dimension χ = d + 1 with O(n) resources, without any fitting, renaming, or self-reference to the target claims. The three-dimensional bond-dimension saturation (∼138, reducible to 39) is obtained by explicit numerical SVD on the Cartesian grid tensor at increasing resolutions, with hardware validation on overlap integrals; this is an empirical observation rather than a reduction to inputs by construction. No load-bearing steps invoke self-citations, fitted parameters renamed as predictions, or ansatzes smuggled via prior work. The derivation chain remains self-contained against external benchmarks.
Axiom & Free-Parameter Ledger
free parameters (1)
- SVD truncation threshold =
10^{-6}
axioms (2)
- domain assumption The amplitude function of an STO on a Cartesian grid admits an MPS representation whose bond dimension saturates with resolution.
- domain assumption Zero-noise extrapolation removes hardware noise sufficiently to validate the ideal MPS integrals at five qubits.
Reference graph
Works this paper leans on
-
[1]
The exponential of a sum factorizes as a product over bits: e−ζx = Y k e−ζ·sk·2n−1−k·∆x.(6) Each factor depends on a single bits k, givingχ= 1 (product state)
Exponential functions The simplest case is a single-sided exponentialf(x) = e−ζx on a gridx= P k sk ·2 n−1−k ·∆xwiths k ∈ {0,1}. The exponential of a sum factorizes as a product over bits: e−ζx = Y k e−ζ·sk·2n−1−k·∆x.(6) Each factor depends on a single bits k, givingχ= 1 (product state). The circuit consists ofnindependent single-qubitR y rotations with a...
-
[2]
A single qubit in superposition controls which branch is active, givingχ= 2
Piecewise exponentials: 1s STO The 1s orbital in one dimension,f(x) =e −ζ|x−R| con- sists of a rising exponential forx < Rand a decaying exponential forx > R. A single qubit in superposition controls which branch is active, givingχ= 2. The circuit 4 is constructed by hand: controlled rotations separate the two branches, each of which is a product state. N...
-
[3]
Linear×exponential: 2s STO The one-dimensional 2s STOf(x) =x·e −ζx hasχ=
-
[4]
The normalizationNis computed inO(n) by contractingχ 2× χ2 = 4×4 transfer matrices
Sincex= P k sk ·2 n−1−k is a sum over bits while e−ζx =Q k e−ζ·sk·2n−1−k is a product, their product has MPS transfer matrices: A[k]0 =I 2, A [k]1 =e k 1w k 0 1 ,(7) wheree k =e −ζ·2n−1−k·∆x andw k = 2 n−1−k/N. The normalizationNis computed inO(n) by contractingχ 2× χ2 = 4×4 transfer matrices
-
[5]
The direct sum givesχ= 3, with 3×3 upper-triangular transfer matrices: A[k]0 =I 3, A [k]1 =e k 1 0 0 0 1−p k/a 0 0 1 ,(8) wheree k =e −2n−1−k/(2a) andp k = 2 n−1−k
Hydrogen 2s orbital The one-dimensional radial partf(r) = (2−r/a)· e−r/2a decomposes as the difference of aχ= 1 term (constant×exponential) and aχ= 2 term (linear× exponential). The direct sum givesχ= 3, with 3×3 upper-triangular transfer matrices: A[k]0 =I 3, A [k]1 =e k 1 0 0 0 1−p k/a 0 0 1 ,(8) wheree k =e −2n−1−k/(2a) andp k = 2 n−1−k. The le...
-
[6]
The physical context in which this function arises is discussed in Sec
Degree-2 polynomial×exponential The functionf(r) =r(2−r/a)·e −r/2a = (2r− r2/a)·e −r/2a is a degree-2 polynomial×exponential withχ= 3. The physical context in which this function arises is discussed in Sec. II D. The three bond states track the running polynomial: state 0 carries the con- stant (1), state 1 the linear sum (x), and state 2 the quadratic su...
-
[7]
The transfer matrices are (d+1)×(d+1) upper triangular, with entries determined by the binomial expansion of (P sk ·2 n−1−k)m form= 0,
General pattern For a one-dimensional function of the formp d(x)·e −ζx wherep d is a polynomial of degreed, the MPS bond dimension isχ=d+ 1. The transfer matrices are (d+1)×(d+1) upper triangular, with entries determined by the binomial expansion of (P sk ·2 n−1−k)m form= 0, . . . , d. All tensors, normalization constants, and circuit parameters are compu...
-
[8]
The angular part is a con- stant and requires no quantum encoding
Spherical coordinates For s-orbitals, the 3D wavefunctionψ(r, θ, φ) =R(r)· Y 0 0 is spherically symmetric. The angular part is a con- stant and requires no quantum encoding. The state is prepared on a single radial register using the 1D analyt- ical MPS constructions of Sec. II C. In spherical coordinates, overlap integrals include the volume elementr 2dr...
-
[9]
The 3D functione −ζ √ x2+y2+z2 is encoded on three registers|x⟩ |y⟩ |z⟩as a single MPS over all 3n qubits via numerical right-canonical SVD (Sec
Cartesian coordinates For multi-center problems, Cartesian coordinates pro- vide a natural grid on which orbitals centered at different positions can be represented without coordinate trans- formations. The 3D functione −ζ √ x2+y2+z2 is encoded on three registers|x⟩ |y⟩ |z⟩as a single MPS over all 3n qubits via numerical right-canonical SVD (Sec. II B). N...
-
[10]
0⟩givesP(0) = | ⟨ψA|ψB⟩ |2
Overlap integral The overlapS AB =⟨ψ A|ψB⟩between two amplitude-encoded orbitals is evaluated using the compute/uncompute method on a single register: ⟨ψA|ψB⟩=⟨0|U † AUB |0⟩.(10) The probability of measuring|0. . .0⟩givesP(0) = | ⟨ψA|ψB⟩ |2. The physical overlap is: SAB =N A · NB ·(± p P(0))·∆x,(11) whereN A,N B are the discrete L2 norms and the sign is d...
-
[11]
This reduces the kinetic energy to an overlap of derivative states
Kinetic energy integral The kinetic energy matrix elementT AB = − 1 2 ⟨ψA| d2 dx2 |ψB⟩is computed using the identity TAB = 1 2 ⟨∂ψA/∂x|∂ψ B/∂x⟩, obtained by integration by parts with vanishing boundary terms. This reduces the kinetic energy to an overlap of derivative states. For the 1s STO, the derivative isd/dx[e −ζ|x−R|] = ±ζ·e −ζ|x−R|, differing from ...
-
[12]
Nuclear attraction integral The nuclear attraction matrix elementV AB = ⟨ψA|(−Z/|x−R C|)|ψ B⟩cannot be evaluated by insert- ing a unitary operator between compute and uncompute steps, because the potentialV(x) =−Z/|x−R C|is a di- agonal but non-unitary operator—it multiplies each basis state by a different real number, changing the norm. Instead, we encod...
-
[13]
0⟩yields P(0) =|S AB|2
Overlap integral The overlap integralS AB =⟨ψ A|ψB⟩is computed via the compute/uncompute circuit: the state|0⟩is evolved to|ψ B⟩by the preparation unitaryU B, thenU † A is ap- plied, and the probability of measuring|0. . .0⟩yields P(0) =|S AB|2. For 1s orbitals, all amplitudes are posi- tive, soS AB = p P(0) without sign ambiguity. Figure 1(a) shows the s...
-
[14]
Integration by parts gives the equivalent formT AB = 1 2 ⟨∂ψA/∂x|∂ψ B/∂x⟩, which reduces the computation to an overlap of derivative states
Kinetic energy integral The kinetic energy matrix elementT AB = − 1 2 ⟨ψA| d2 dx2 |ψB⟩is evaluated using the derivative encoding of the 1s orbital. Integration by parts gives the equivalent formT AB = 1 2 ⟨∂ψA/∂x|∂ψ B/∂x⟩, which reduces the computation to an overlap of derivative states. Sinced/dx[e −ζ|x−R|] =±ζ·e −ζ|x−R| with a sign reversal at the cente...
-
[15]
Nuclear attraction integral The nuclear attraction matrix elementV AB = ⟨ψA|(−Z/|x−R C|)|ψ B⟩is computed by encoding the product functionφ B(x) =V(x)·ψ B(x) as a separate MPS state and reducing the matrix element to an over- lap:V AB =N A · Nφ · ⟨ψnorm A |φnorm B ⟩ ·∆x, whereN A and Nφ are the discrete L2 norms computed classically during MPS construction...
-
[16]
Spherical coordinates: same-center orthogonality In spherical coordinates, the radial parts of 1s and 2s orbitals are encoded on a single register using the analyt- ical MPS constructions described in Sec. II C. Since the angular parts are identical for s-orbitals, the orthogonal- ity⟨1s|2s⟩is determined entirely by the radial overlap. We simulate this on...
-
[17]
Cartesian coordinates: multi-center overlaps For the Cartesian representation, 1s and 2s orbitals are encoded on three registers|x⟩ |y⟩ |z⟩using MPS decompo- sition over all 3nqubits (Sec. II B). The simulated space FIG. 2. 3D overlap convergence. (a)⟨1s A|1sB⟩atd= 1.4 a.u. (b)⟨2s A|2sB⟩atd= 1.4 a.u. Dotted line: exact continuous value. spans 16 to 64 a.u...
-
[18]
The 1s orbitale −ζ|x−R| hasχ= 2, with the bond state distinguishing the rising and decay- ing branches on either side of the center
Bond dimensions of 1D orbital encodings Table I summarizes the MPS bond dimensions for all 1D functions studied. The 1s orbitale −ζ|x−R| hasχ= 2, with the bond state distinguishing the rising and decay- ing branches on either side of the center. For polynomial ×exponential functions of the formp(x)·e −ζx, the bond dimension equalsd+ 1 wheredis the polynom...
-
[19]
We study the scaling of both quantities under three variables: grid resolution (qubits per coordinate), orbital decay constant ζ, and SVD truncation threshold
Bond dimensions of 3D Cartesian encodings The entanglement structure of three-dimensional or- bitals encoded on a Cartesian grid|x⟩ |y⟩ |z⟩is charac- terized by two quantities: the maximum bond dimen- sionχ max along the MPS chain, and the cross-coordinate Schmidt rankχ(x|y) at the register boundary. We study the scaling of both quantities under three var...
-
[20]
Bond profile structure Figure 4 shows the full bond dimension profile across the 3n-qubit MPS chain for the 1s orbital at 11 qubits per coordinate. The profile reveals a characteristic arch shape: bonds within theyregister reach the maximum (χ= 132 at threshold 10 −12), while bonds within thex andzregisters peak at approximately 58. This asym- metry is no...
-
[21]
Qubit ordering We compared the grouped ordering|x⟩ |y⟩ |z⟩with the interleaved ordering|x 1y1z1x2y2z2 . . . xnynzn⟩. At 3 qubits per coordinate, the interleaved ordering pro- duces a maximum bond dimension of 64 versus 28 for grouped, and the transpiled circuit contains 39,026 ECR gates versus 9,552—a factor of 4 increase (Table II). At largernthe interle...
-
[22]
Two-electron integrals The Coulomb interaction kernel 1/|x 1 −x 2|on two registers has full Schmidt rank (equal to the grid size N) across thex 1|x2 bipartition. The full integrand f1(x1)·(1/|x1 −x2|)·f2(x2), despite the exponential local- ization of the orbital charge distributions, retains near- full Schmidt rank across thex 1|x2 bipartition at all grid...
-
[23]
5; Table V]
Overlap integral on hardware The best hardware result is the 5-qubit 1s overlap on ibm kingston with ZNE, yieldingS AB = 0.576 versus the statevector value of 0.572—a hardware-induced error of 0.67% [Fig. 5; Table V]. The total error relative to the exact continuous integral is 3.2%, dominated by the 2.5% discretization error inherent to the 5-qubit grid....
-
[24]
5) versus the statevector value−0.0628 (9.6% hardware noise)
Kinetic energy integral on hardware The kinetic energy integral was measured on ibm kingston at 5 qubits with ZNE, yieldingT AB = −0.0568 (Fig. 5) versus the statevector value−0.0628 (9.6% hardware noise). The larger relative noise com- pared to the overlap arises from the small absolute value of the matrix element: the probabilityP(0) =|T AB|2 ≈ 0.004 si...
-
[25]
efficient
Circuits beyond current hardware The 3D orthogonality circuit⟨1s|2s⟩in spherical co- ordinates requires 149–310 ECR gates at 4–8 qubits per coordinate (Table II). While these circuits are more com- pact than the Cartesian 3D encodings, noisy simulation on thefake sherbrookenoise model shows that hard- 12 FIG. 5. Hardware validation at 5 qubits on ibm king...
-
[26]
J. C. Slater, Phys. Rev.36, 57 (1930)
1930
-
[27]
W. J. Hehre, R. F. Stewart, and J. A. Pople, J. Chem. Phys.51, 2657 (1969)
1969
-
[28]
Szabo and N
A. Szabo and N. S. Ostlund,Modern Quantum Chem- istry: Introduction to Advanced Electronic Structure The- ory(Dover, New York, 1996)
1996
-
[29]
V. V. Shende, S. S. Bullock, and I. L. Markov, IEEE Trans. CAD25, 1000 (2006)
2006
-
[30]
Peruzzoet al., Nat
A. Peruzzoet al., Nat. Commun.5, 4213 (2014)
2014
-
[31]
Babbushet al., Phys
R. Babbushet al., Phys. Rev. X8, 041015 (2018)
2018
-
[32]
H¨ aner, M
T. H¨ aner, M. Roetteler, and K. M. Svore, Quantum Sci. Technol.3, 035005 (2018)
2018
-
[33]
J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, and H. Neven, Nat. Commun.9, 4812 (2018)
2018
-
[34]
Sch¨ on, E
C. Sch¨ on, E. Solano, F. Verstraete, J. I. Cirac, and M. M. Wolf, Phys. Rev. Lett.95, 110503 (2005)
2005
-
[35]
Ran, Phys
S.-J. Ran, Phys. Rev. A101, 032310 (2020)
2020
-
[36]
M. A. Nielsen and I. L. Chuang,Quantum Computation and Quantum Information, 10th Anniversary ed. (Cam- bridge University Press, Cambridge, 2010)
2010
-
[37]
Brassard, P
G. Brassard, P. Høyer, M. Mosca, and A. Tapp, Con- temp. Math.305, 53 (2002)
2002
-
[38]
Schollw¨ ock, Ann
U. Schollw¨ ock, Ann. Phys.326, 96 (2011)
2011
-
[39]
I. V. Oseledets and E. E. Tyrtyshnikov, SIAM J. Sci. Comput.33, 2295 (2011)
2011
-
[40]
IBM Quantum Platform, Processor types, https://quantum.cloud.ibm.com/docs/en/guides/processor- types (accessed April 2026)
2026
-
[41]
https://doi.org/10.5281/zenodo.2573505
Qiskit contributors, “Qiskit: An open-source framework for quantum computing” (2023), doi:10.5281/zenodo.2573505
-
[42]
Temme, S
K. Temme, S. Bravyi, and J. M. Gambetta, Phys. Rev. Lett.119, 180509 (2017)
2017
-
[43]
Li and S
Y. Li and S. C. Benjamin, Phys. Rev. X7, 021050 (2017)
2017
-
[44]
M. P. Barnett and C. A. Coulson, Phil. Trans. R. Soc. Lond. A243, 221 (1951)
1951
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.