Nielsen's Geometric Approach to Complexity

• Nielsen's Geometric Approach is a framework that reformulates complexity problems as minimal geodesic lengths on Lie-group manifolds.
• It applies to quantum circuit complexity, topological coincidence theory, and open-system dynamics by translating combinatorial issues into geometric terms.
• The method transfers minimality questions to an auxiliary geometric space, enabling precise evaluations of complexity via invariant metrics and exact geodesic solutions.

Nielsen's geometric approach denotes a class of constructions in which a problem is reformulated in geometric terms so that the relevant invariant is expressed through geodesic length, graph intersection classes, or bordism data. In the quantum-information and quantum-field-theoretic literature, it recasts circuit complexity as the minimal length of a geodesic on a Lie-group manifold of allowed operators, equipped with a right-invariant metric (Chowdhury et al., 16 Dec 2025). In topological coincidence theory, a geometric Nielsen construction analyzes coincidences through graph intersections in $X\times Y$ and through path-space or bordism decompositions of the coincidence locus (Graff et al., 7 May 2026). This suggests a common structural theme: minimality questions are transferred from a direct combinatorial or pointwise description to an auxiliary geometric space.

1. Quantum circuit complexity as a geodesic problem

In the quantum formulation, a circuit implementing a target unitary $U_T$ is represented by a path-ordered exponential

$$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$

with $U(0)=\mathbb{1}$ and $U(1)=U_T$. The cost of a path is written as

$$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$

and the circuit complexity is the minimum of this cost over admissible paths. In the Lie-algebraic formulation, the same quantity is expressed as

$$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$

where $G_{IJ}$ is the metric in the chosen generator basis and may impose penalty factors for “hard” directions (Bernamonti et al., 2019, Chowdhury et al., 16 Dec 2025).

This framework treats the set of allowed unitary operators as a Lie group manifold. Quantum operations are points on that manifold, and quantum computations are trajectories generated by time-dependent Hamiltonians. For $SU(N)$, a right-invariant Riemannian metric can be written as

$$\langle \hat A,\hat B\rangle_\Omega = \frac{1}{N^2-1}\,\vec A^\dagger \Omega \vec B,$$

and the associated intrinsic distance is the Carnot-Carathéodory metric

$U_T$0

The geometric complexity is then $U_T$1 (Acevedo et al., 24 Jul 2025).

The formalism does not fix a unique cost function. The literature summarized here emphasizes smoothness, positive-definiteness, the triangle inequality, and positive homogeneity of degree one in tangent vectors as natural criteria for $U_T$2. Different choices of metric or cost can therefore lead to different complexity assignments, even for the same reference and target data (Bernamonti et al., 2019).

2. Metric structure, generators, and exact geodesics

A central technical ingredient is the choice of generators and the associated metric. For Gaussian systems, the relevant operators often close into a finite-dimensional Lie algebra, and the geometry becomes Riemannian. In the $U_T$3 case relevant for quantum scalar fields on homogeneous and isotropic cosmological backgrounds, the generators are

$U_T$4

$U_T$5

$U_T$6

with geodesics governed by the Euler-Arnold equations

$U_T$7

Using a finite-dimensional faithful matrix representation, the geodesic problem can be solved exactly instead of by perturbative Dyson-series or Baker-Campbell-Hausdorff truncations. The resulting exact complexity for a general $U_T$8 unitary is

$U_T$9

and for a pure two-mode squeezing operator one finds

$$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$0

(Chowdhury et al., 16 Dec 2025).

In lattice Proca theory, the same geometric program is implemented with the quadratic cost

$$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$1

which induces the right-invariant metric

$$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$2

After discretization and suitable Fourier and linear transformations, the theory reduces to decoupled harmonic oscillators. The ground and reference states are Gaussian and are described by covariance matrices $$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$3 and $$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$4, related by $$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$5 with

$$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$6

For the $$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$7 cost, the minimal path is a straight line in the group manifold (Meng et al., 2021).

These developments show that the computational content of Nielsen's approach depends heavily on the algebraic realization of the gate set. When a faithful matrix representation is available, exact geodesic distance can replace upper bounds; when the target states are Gaussian, covariance-matrix methods make the geometry explicitly calculable.

3. Applications in many-body physics and quantum field theory

The approach has been used extensively in free and integrable models. In the one-dimensional Kitaev chain, the ground state factorizes over momentum pairs $$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$8, and the complexity between two ground states is

$$U(s)=\mathcal{P}\exp\left(-i\int_0^s H(s')\,ds'\right), \qquad H(s)=\sum_I Y^I(s)\,\mathcal{O}_I,$$9

or, in the thermodynamic limit,

$U(0)=\mathbb{1}$0

The derivative of the complexity with respect to the target parameter develops a logarithmic divergence at the critical point, and the real-space optimal Hamiltonian is effectively local when the two states lie in the same phase but intrinsically non-local when they lie in different phases (Liu et al., 2019).

For coherent states in free scalar field theory, the gate set must be enlarged to include shift gates in addition to scaling and entangling gates. The relevant group becomes $U(0)=\mathbb{1}$1. A notable feature is that optimal circuits can introduce entanglement between normal modes at intermediate stages even when both the reference and target states are unentangled in that basis. The vacuum complexity is UV divergent, but the finite increase over the vacuum for coherent states is UV finite for $U(0)=\mathbb{1}$2 and for Schatten norms with $U(0)=\mathbb{1}$3. For general coherent states, both the complexity values and the optimal circuits differ from those obtained with the Fubini-Study method (Guo et al., 2018).

In Proca theory, the ground-state complexity receives contributions from both $U(0)=\mathbb{1}$4- and $U(0)=\mathbb{1}$5-components, and the thermofield double state exhibits a late-time behavior

$U(0)=\mathbb{1}$6

By contrast, the Fubini-Study metric yields linear growth in time for the thermofield double state. The summary provided for this model attributes the difference to the extra longitudinal mode and to the metric dependence of the construction (Meng et al., 2021).

In quantum cosmology, the exact $U(0)=\mathbb{1}$7 treatment gives a de Sitter expression

$U(0)=\mathbb{1}$8

while in asymptotically static universes the complexity between “in” and “out” vacua reduces to

$U(0)=\mathbb{1}$9

These formulas retain phase-angle dependence that perturbative upper-bound methods miss (Chowdhury et al., 16 Dec 2025).

4. Endpoint variation, conformal circuits, and holography

A general variational statement derived within the geometric framework is the first law of complexity. If $U(1)=U_T$0 denotes the optimal trajectory and the target state is varied so that the endpoint shifts by $U(1)=U_T$1, then the first-order variation of the complexity is

$U(1)=U_T$2

If the endpoint variation is orthogonal to the tangent at the endpoint, the first-order term vanishes and the second-order law becomes

$U(1)=U_T$3

The defining feature is that the variation depends only on the endpoint of the optimal circuit and not on the bulk history of the path (Bernamonti et al., 2019).

In two-dimensional conformal field theories, circuits can be built from conformal transformations, so the relevant paths lie in the Virasoro group. The complexity functional takes the form

$U(1)=U_T$4

and this functional is equivalent to the Polyakov action of two-dimensional gravity, or, equivalently, to the geometric action on Virasoro coadjoint orbits (Caputa et al., 2018).

The holographic application of the first law uses the complexity=action proposal. For a coherent scalar perturbation of AdS, all gravitational contributions to the variation of the Wheeler-DeWitt action cancel once the appropriate null-boundary counterterm is included, and the surviving result is

$U(1)=U_T$5

The null boundary of the Wheeler-DeWitt patch is therefore interpreted as acting like the “end of the quantum circuit” (Bernamonti et al., 2019).

5. Open-system extensions and quantum channels

A direct extension of Nielsen's construction to open dynamics is obstructed by the fact that the set of quantum channels is not a Lie group and does not carry the same canonical manifold structure as $U(1)=U_T$6. One route is purification: represent the channel by a unitary evolution on system plus environment and then apply geometric complexity to the dilation. In the Hilbert-Schmidt geometry, the lifted unitary $U(1)=U_T$7 has complexity

$U(1)=U_T$8

and an improved channel complexity is defined by subtracting the part deemed invisible to the system,

$U(1)=U_T$9

The corresponding noise complexity is the absolute difference between the channel complexity and the system-only unitary complexity (Acevedo et al., 24 Jul 2025).

A more explicit formulation introduces a dilation-based functional for a specific Stinespring realization $$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$0,

$$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$1

together with an intrinsic channel complexity obtained by minimizing over an admissible class of dilations,

$$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$2

The subtraction is justified by closed-system consistency, environment-only neutrality, stability under environment-gauge transformations, and a variational principle selecting the canonical surrogate Hamiltonian

$$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$3

For time-independent dilations, the implementation-dependent complexity scales linearly in time,

$$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$4

and in the GKSL/Lindblad regime there are dissipator-controlled upper bounds involving $$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$5. The framework is illustrated on dephasing, amplitude damping, and depolarizing channels (Acevedo et al., 2 Jan 2026).

These channel constructions preserve the central geometric intuition of the original unitary theory while making explicit that, for open dynamics, the definition of complexity depends on what microscopic resources are counted and what dilation freedoms are regarded as gauge.

6. Geometric Nielsen theory in coincidence topology

In topological coincidence theory, the geometric approach focuses on pairs $$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$6 in which $$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$7 is an $$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$8-valued map and $$D[x^a(s)] = \int_0^1 ds\, F\!\left(x^a,\dot x^a\right),$$9 is an $$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$0-valued map on connected finite polyhedra. A point $$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$1 is a coincidence point if

$$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$2

The domain coincidence set is

$$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$3

while the graph-intersection set is

$$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$4

The revision of the 2013 Brown-Kolahi construction proceeds by treating coincidences as intersections of the graphs

$$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$5

rather than as points in the domain alone. Two graph-intersection points are declared equivalent when they are related by a path in $$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$6 along which appropriate branches of splittings of $$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$7 and $$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$8 match at the endpoints and are homotopic relative to endpoints (Graff et al., 7 May 2026).

The resulting invariant counts geometrically essential graph-intersection classes: $$C[U_\mathrm{target}] = \min_{\{V^I(s)\}} \int_0^1 ds \sqrt{G_{IJ}V^I(s)V^J(s)},$$9 For the circle, an algebraic version counts classes of nonzero local intersection index, and the main result is that for an $G_{IJ}$0-valued map of degree $G_{IJ}$1 and an $G_{IJ}$2-valued map of degree $G_{IJ}$3,

$G_{IJ}$4

Moreover, there exists a homotopic pair with exactly this number of graph intersection points in $G_{IJ}$5. The earlier Brown-Kolahi formula

$G_{IJ}$6

can overestimate the minimal number of coincidence points because the equivalence relation on domain coincidences is not transitive and does not capture the correct homotopy classes of coincidences (Graff et al., 7 May 2026).

A related but broader geometric framework for coincidences of single-valued maps $G_{IJ}$7 uses the path-space fibration

$G_{IJ}$8

and partitions the coincidence set into Nielsen classes indexed by path components of $G_{IJ}$9. From the coincidence manifold and its normal data, one obtains four Nielsen numbers: $SU(N)$0, $SU(N)$1, $SU(N)$2, and $SU(N)$3. They satisfy

$SU(N)$4

and, in the fixed-point setting, all four coincide with the classical Nielsen number. In higher codimension they can differ, and explicit computations for maps from spheres to real, complex, or quaternionic projective spaces yield both Wecken-type equalities and counterexamples detected by Kervaire invariants (Koschorke, 2013).

The topological literature therefore uses “geometric Nielsen” in a sense parallel to, but distinct from, quantum complexity: the essential object is not a geodesic in operator space but a decomposition of a coincidence phenomenon into geometrically meaningful classes whose nontriviality survives homotopy.