Fundamental Gap Algorithm (FGA)

• Fundamental Gap Algorithm (FGA) is a quantum algorithm designed to compute the ground state energy of convex Schrödinger operators using adiabatic evolution and phase estimation.
• It overcomes the curse of dimensionality with a low-energy truncation framework and spectral gap estimates, ensuring polynomial runtime in key parameters.
• Its modular design bridges convex optimization, spectral geometry, and quantum computing, inspiring future solutions for high-dimensional variational problems.

Searching arXiv for the specific papers and closely related foundational work on the Fundamental Gap Algorithm and the fundamental gap theorem. The Fundamental Gap Algorithm (FGA) is a quantum algorithm for the quantum analogue of unconstrained convex optimization: computing the minimum eigenvalue of a Schrödinger operator

$h=-\Delta+V$

with convex potential $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$ such that $V(x)\rightarrow\infty$ as $\|x\|\rightarrow\infty$. Its target quantity is the ground state energy

$$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$

and the algorithm computes this value up to error $\epsilon$ in polynomial time in $n$, $1/\epsilon$, and parameters that depend on $V$ (Lee, 2 Oct 2025). The construction combines discretization, adiabatic evolution of the ground state, phase estimation, and a low-energy truncation framework. Its analysis is tied to lower bounds on the fundamental gap, i.e. the separation between the two lowest eigenvalues, which is the spectral parameter governing adiabatic performance.

1. Problem class and optimization viewpoint

The FGA is formulated for Schrödinger operators on $\mathbb R^n$ with convex confining potentials. In the source formulation, this is presented as a quantum analogue of convex optimization: instead of minimizing a convex scalar function over $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$0, the task is to compute the minimum eigenvalue of a quantum Hamiltonian $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$1 under the assumptions that $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$2 is convex, nonnegative, and diverges at infinity (Lee, 2 Oct 2025).

This perspective is technically significant because the objective is inherently quantum. The kinetic term $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$3 prevents reduction to ordinary classical convex minimization, and the variational problem is over normalized wavefunctions rather than points in configuration space. The paper explicitly contrasts this with approaches that quantize classical optimization dynamics: FGA directly targets a quantum mechanical energy functional rather than a classical convex objective (Lee, 2 Oct 2025).

The motivation is computational. The underlying eigenvalue problem is a high-dimensional PDE/spectral problem, and the source summary states that classical methods suffer from the curse of dimensionality, while the FGA attains polynomial scaling for the stated class of Hamiltonians (Lee, 2 Oct 2025). A plausible implication is that the algorithm is meant to serve as a bridge between convex geometry, spectral theory, and quantum algorithm design.

2. Spectral-geometric foundation

The algorithm rests on the theory of the fundamental gap. For Schrödinger operators with Dirichlet boundary data on convex domains in Euclidean space, Andrews and Clutterbuck proved the Fundamental Gap Conjecture: if $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$4 is a bounded convex domain of diameter $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$5 and $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$6 is convex, then

$V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$7

with equality for the interval $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$8 with zero potential (Andrews et al., 2010). More generally, their proof gives a comparison theorem with a one-dimensional model potential $V:\mathbb R^n\rightarrow \mathbb R_{\ge 0}$9, yielding

$V(x)\rightarrow\infty$0

whenever $V(x)\rightarrow\infty$1 admits the appropriate modulus of convexity (Andrews et al., 2010).

A second ingredient in this line of work is the sharp spherical extension. For a bounded, strictly convex domain $V(x)\rightarrow\infty$2 with Dirichlet Laplacian eigenvalues $V(x)\rightarrow\infty$3, the fundamental gap is

$V(x)\rightarrow\infty$4

For diameter $V(x)\rightarrow\infty$5, Seto, Wang, and Wei proved

$V(x)\rightarrow\infty$6

where $V(x)\rightarrow\infty$7 are the first two Dirichlet eigenvalues of the one-dimensional spherical model operator

$V(x)\rightarrow\infty$8

and for $V(x)\rightarrow\infty$9, $\|x\|\rightarrow\infty$0,

$\|x\|\rightarrow\infty$1

The proof adapts the Andrews–Clutterbuck strategy through super log-concavity of the first eigenfunction, two-point-function maximum principles, and parabolic arguments (Seto et al., 2016).

Within the FGA framework, the role of these results is structural rather than merely historical. The 2025 paper identifies the lower bound

$\|x\|\rightarrow\infty$2

as the critical spectral-gap estimate used in the adiabatic analysis, with $\|x\|\rightarrow\infty$3 denoting the diameter of the relevant domain in the convex setting (Lee, 2 Oct 2025). This ties the algorithm directly to the geometric spectral theory developed in the fundamental-gap literature.

3. Algorithmic construction

The algorithm begins by discretizing the continuous operator on a finite grid within a torus $\|x\|\rightarrow\infty$4, producing a finite Hamiltonian $\|x\|\rightarrow\infty$5 that mimics $\|x\|\rightarrow\infty$6 with exponentially small error (Lee, 2 Oct 2025). In the stated construction, the kinetic term is diagonal in the Fourier basis and the potential term is diagonal in the computational basis.

The core computational primitive is adiabatic evolution. The paper uses a time-dependent Hamiltonian

$\|x\|\rightarrow\infty$7

where $\|x\|\rightarrow\infty$8 is the discretized Laplacian, $\|x\|\rightarrow\infty$9 is an initial bowl-shaped potential, and $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$0 is the discretized target potential, possibly with saturation or cutoff for technical reasons (Lee, 2 Oct 2025). The initial ground state is chosen so that it is easy to prepare; the source summary describes it as a product of single-qudit states. After state preparation, the algorithm simulates the adiabatic path and then applies phase estimation to the final Hamiltonian to read out an estimate of the ground state energy (Lee, 2 Oct 2025).

In the step-by-step outline given in the source summary, the procedure is:

1. prepare the ground state of the simple initial Hamiltonian $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$1;
2. use phase estimation and selection for initialization;
3. simulate the adiabatic evolution from $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$2 to $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$3, with truncated Dyson series mentioned as one simulation technique;
4. apply phase estimation to the final state with respect to $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$4;
5. output the measured eigenvalue as an estimate of $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$5 (Lee, 2 Oct 2025).

The construction is therefore not a direct eigensolver in the classical numerical-analysis sense. It is an adiabatic quantum algorithm whose efficiency depends on preserving fidelity in the low-energy sector while maintaining control over the spectral gap.

4. Low-energy truncation and correctness analysis

A distinguishing technical innovation of the FGA is its low-energy truncation framework. The paper’s central insight is that only the low-energy subspace matters for adiabatic success, so high-energy misbehaviors can be ignored or controlled (Lee, 2 Oct 2025). This is used to move rigorously between several representations of the problem: the original infinite-dimensional Hamiltonian, spatially truncated versions, the torus-discretized operator, and the eventual qubit implementation.

The source summary describes a modular equivalence theory for these Hamiltonians. If two self-adjoint Hamiltonians agree up to $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$6 in their $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$7-truncated domain, then their ground and first excited eigenvalues differ by $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$8; elsewhere in the summary this is sharpened to the statement that if two Hamiltonians are $$\lambda_0=\min_{\langle \psi\mid\psi\rangle=1}\langle \psi\mid h\mid \psi\rangle,$$9-equivalent, then their ground energies are within $\epsilon$0 (Lee, 2 Oct 2025). This is the formal device that justifies replacing the continuous operator by a simulable finite model without losing control of the low spectrum relevant to adiabatic evolution.

The main performance statement is that FGA computes the ground state energy to error $\epsilon$1 in time polynomial in $\epsilon$2, $\epsilon$3, and parameters depending on $\epsilon$4 (Lee, 2 Oct 2025). The runtime is summarized as

$\epsilon$5

where the geometry parameters encode domain size, diameter, and level-set information, and the smoothness parameters encode derivative bounds and the efficiency of quantum circuit access to $\epsilon$6 (Lee, 2 Oct 2025). The same source emphasizes that all crucial steps—state preparation, adiabatic evolution, and measurement—are polynomial under the stated assumptions.

This analytical organization has broader significance. The paper explicitly presents low-energy truncation as a precise framework for ignoring high-energy subspaces in infinite-dimensional settings, and its modular adiabatic analysis is one of the principal methodological contributions of the work (Lee, 2 Oct 2025).

5. Application to the convex drum problem

A major application given for the FGA is the Dirichlet Laplacian on polyhedral domains, described in the paper as the problem of finding the lowest frequency of an $\epsilon$7-dimensional convex drum (Lee, 2 Oct 2025). The source abstract states that the algorithm gives the first known polynomial-time method for computing the minimum eigenvalue of the Dirichlet Laplacian on an $\epsilon$8-dimensional region defined by $\epsilon$9 linear constraints, with runtime polynomial in $n$0, $n$1, $n$2, and the radius $n$3 of a ball containing the region (Lee, 2 Oct 2025).

The technical mechanism is a smooth barrier construction. For a polytope

$n$4

the paper introduces a penalty potential of the form

$n$5

Inside the region, $n$6; outside, $n$7 grows rapidly (Lee, 2 Oct 2025). In this way, Dirichlet boundary conditions are enforced softly by penalizing amplitude outside $n$8, and the truncation and discretization analysis continues to apply.

This application is important because it translates a geometric PDE eigenvalue problem into the Schrödinger-operator framework already handled by the FGA. A plausible implication is that the algorithm is not limited to abstract convex potentials on $n$9, but also serves as a spectral algorithm for high-dimensional convex geometry once boundary conditions are encoded by smooth penalties.

6. Terminology, related methods, and scope

The term “FGA” requires terminological care. In the broader literature represented here, the same acronym also denotes the Frozen Gaussian Approximation, a semiclassical wave-propagation method, and the related Frozen Gaussian Grid-point Correction, a reconstruction algorithm for the semiclassical Schrödinger equation (Chai et al., 30 Apr 2025, Hateley et al., 2018). Those works concern Gaussian wave-packet decompositions, ray-based propagation, least-squares on-grid correction, and FFT-accelerated reconstruction, rather than the spectral-gap-based quantum optimization algorithm denoted Fundamental Gap Algorithm. The overlap is therefore lexical, not methodological.

Within quantum algorithms, the FGA is positioned as a method for optimizing a quantum objective rather than a classical convex function. The source summary states that, unlike approaches that quantize classical optimization dynamics, the FGA directly tackles minimization of quantum mechanical energy (Lee, 2 Oct 2025). It is also presented as offering exponential quantum speed-up relative to classical methods for the stated class of high-dimensional spectral problems, under standard complexity assumptions (Lee, 2 Oct 2025).

The paper also assigns broader significance to its technical framework. It states that low-energy truncation and modular adiabatic analysis may inspire future quantum algorithms for PDEs, quantum chemistry, and other variational problems in high dimension (Lee, 2 Oct 2025). At the same time, its complexity-theoretic implications remain open-ended: BQP-completeness for convex-quantum Hamiltonian problems is described as speculated but not yet proven (Lee, 2 Oct 2025). This suggests that the FGA is both a concrete algorithmic result and a template for a wider research program linking convexity, spectral geometry, and quantum complexity.