Paper ReviewMathematics & StatisticsOptimization & Operations Research
Quantum Gates on Curved Manifolds: Sub-Riemannian Geometry for Energetically Optimal Computation
Implementing a quantum gate requires driving a quantum system along a path through the special unitary group SU(n). The energetically optimal path is a geodesicβbut in the sub-Riemannian geometry where only certain control directions are available, geodesics are fundamentally different from Riemannian ones.
By Sean K.S. Shin
This blog summarizes research trends based on published paper abstracts. Specific numbers or findings may contain inaccuracies. For scholarly rigor, always consult the original papers cited in each post.
Every quantum computation reduces to a sequence of quantum gatesβunitary transformations that evolve the quantum state. Implementing a gate physically requires applying control fields (electromagnetic pulses, laser beams, voltage sequences) that drive the quantum system from its initial state to a target state. The path through state space matters: different control sequences reach the same target but with different energy costs, different durations, and different sensitivities to noise.
The optimal control problemβfinding the control sequence that minimizes energy while reaching the target gateβis naturally formulated in the language of sub-Riemannian geometry. A quantum gate is an element of the special unitary group SU(n). The control Hamiltonian determines which directions in SU(n) are directly accessible (the "horizontal" directions in sub-Riemannian terminology). An energetically optimal control sequence corresponds to a geodesic in this sub-Riemannian structure.
Da Silva et al. develop a Monte Carlo method for finding these geodesicsβcombining the geometric rigor of sub-Riemannian analysis with the computational flexibility of stochastic search.
Why Sub-Riemannian, Not Riemannian?
In Riemannian geometry, you can move in any direction at any point. The geodesic (shortest path) between two points is found by variational calculusβa well-understood problem with efficient computational methods.
In sub-Riemannian geometry, you can move only in certain directionsβthe "horizontal" directions determined by a distribution (a sub-bundle of the tangent bundle). The physical constraint is that quantum systems can only be controlled through a limited set of control Hamiltonians. You cannot directly rotate the quantum state in an arbitrary direction; you can only apply the control fields available in your experimental apparatus.
The sub-Riemannian geodesic problem is fundamentally harder than the Riemannian one:
- Geodesics may not exist between all pairs of points (though the Chow-Rashevskii theorem guarantees reachability for bracket-generating distributions)
- When geodesics exist, they may be abnormalβsatisfying the geodesic equations in a way that has no Riemannian analogue
- The geodesic equations involve the full Lie bracket structure of the control algebra, making them analytically intractable for most systems
The Monte Carlo Geodesic Search
Da Silva et al.'s approach bypasses the analytical intractability by using a Monte Carlo search over the space of control sequences. The algorithm:
Parameterize control sequences as piecewise-constant profiles (a sequence of control amplitudes applied for specified durations)
Evaluate each candidate by computing the resulting unitary transformation and its energy cost
Search the parameter space using Monte Carlo sampling, guided by a cost function that balances gate fidelity (how close the result is to the target gate) against energy expenditure
Refine promising candidates using local optimization (gradient descent on the smooth manifold SU(n))The geometric insight that makes this tractable: the energy cost of a control sequence is directly related to the sub-Riemannian length of the corresponding path in SU(n). Minimizing energy is equivalent to finding the shortest sub-Riemannian pathβa geodesic. The Monte Carlo search explores this geodesic landscape stochastically, finding near-optimal paths that analytical methods cannot reach for realistic quantum systems.
Claims and Evidence
<
| Claim | Evidence | Verdict |
|---|
| Quantum gate optimization is a sub-Riemannian geodesic problem | Mathematical formulation is established | β
Well-established |
| Monte Carlo search finds near-optimal quantum gates | Da Silva et al. demonstrate on small quantum systems | β
Supported |
| Sub-Riemannian geodesics are qualitatively different from Riemannian ones | Abnormal geodesics, bracket-generating constraints | β
Mathematical fact |
| The approach scales to large quantum systems | Computational cost grows with system size; demonstrated on small systems | β οΈ Scaling is a challenge |
Open Questions
Scalability: Current demonstrations involve small quantum systems (2-3 qubits). Can the Monte Carlo geodesic search scale to the multi-qubit gates needed for practical quantum computing?Noise robustness: Optimal gates found in the noiseless sub-Riemannian setting may be sensitive to noise. Can the geometric framework incorporate noise models to find gates that are both energetically optimal and noise-robust?Topological obstructions: The topology of SU(n) (particularly its fundamental group) may create obstructions to certain gate sequences. How does the sub-Riemannian structure interact with these topological constraints?Connection to quantum error correction: Quantum error correction codes define subspaces of the full Hilbert space. Can sub-Riemannian optimization find gates that operate within the code space while minimizing energy?What This Means for Your Research
For quantum computing researchers, sub-Riemannian optimization provides a principled framework for pulse designβreplacing ad hoc optimization of control parameters with geometrically motivated search over the natural manifold of quantum gates.
For differential geometers, quantum computing provides physically motivated sub-Riemannian problems on Lie groupsβa class of problems where the geometry is rich and the applications are immediate.
λ©΄μ±
μ‘°ν: μ΄ κ²μλ¬Όμ μ 보 μ 곡μ μν μ°κ΅¬ λν₯ κ°μμ΄λ€. νΉμ μ°κ΅¬ κ²°κ³Ό, ν΅κ³, μ£Όμ₯μ νμ μ°κ΅¬μμ μΈμ©νκΈ° μ μ μλ¬Έ λ
Όλ¬Έμ ν΅ν΄ κ²μ¦ν΄μΌ νλ€.
곑λ₯ λ€μ체 μμ μμ κ²μ΄νΈ: μλμ§ μ΅μ κ³μ°μ μν μ€-λ¦¬λ§ κΈ°νν
λͺ¨λ μμ κ³μ°μ μμ μνλ₯Ό μ§νμν€λ μ λν°λ¦¬ λ³νμΈ μμ κ²μ΄νΈμ μνμ€λ‘ νμλλ€. κ²μ΄νΈλ₯Ό 물리μ μΌλ‘ ꡬννλ €λ©΄ μμ μμ€ν
μ μ΄κΈ° μνμμ λͺ©ν μνλ‘ μ λνλ μ μ΄μ₯(μ μκΈ° νμ€, λ μ΄μ λΉ, μ μ μνμ€)μ μΈκ°ν΄μΌ νλ€. μν 곡κ°μ ν΅κ³Όνλ κ²½λ‘κ° μ€μνλ€: μλ‘ λ€λ₯Έ μ μ΄ μνμ€κ° λμΌν λͺ©νμ λλ¬νλλΌλ μλμ§ λΉμ©, μμ μκ°, μ‘μμ λν λ―Όκ°λλ κ°κ° λ€λ₯΄λ€.
μλμ§λ₯Ό μ΅μννλ©΄μ λͺ©ν κ²μ΄νΈμ λλ¬νλ μ μ΄ μνμ€λ₯Ό μ°Ύλ μ΅μ μ μ΄ λ¬Έμ λ μ€-λ¦¬λ§ κΈ°νν(sub-Riemannian geometry)μ μΈμ΄λ‘ μμ°μ€λ½κ² μ μνλλ€. μμ κ²μ΄νΈλ νΉμ μ λν°λ¦¬ κ΅° SU(n)μ μμμ΄λ€. μ μ΄ ν΄λ°ν λμμ SU(n) λ΄μμ μ§μ μ κ·Ό κ°λ₯ν λ°©ν₯(μ€-λ¦¬λ§ μ©μ΄λ‘ "μν" λ°©ν₯)μ κ²°μ νλ€. μλμ§ μ΅μ μ μ΄ μνμ€λ μ΄ μ€-λ¦¬λ§ κ΅¬μ‘° μμμμ μΈ‘μ§μ (geodesic)μ ν΄λΉνλ€.
Da Silva λ±μ μ΄ μΈ‘μ§μ μ μ°ΎκΈ° μν λͺ¬ν
μΉ΄λ₯Όλ‘ λ°©λ²μ κ°λ°νμμΌλ©°, μ΄λ μ€-λ¦¬λ§ ν΄μμ κΈ°ννμ μλ°μ±κ³Ό νλ₯ μ νμμ κ³μ°μ μ μ°μ±μ κ²°ν©ν κ²μ΄λ€.
μ 리λ§μ΄ μλ μ€-리λ§μΈκ°?
λ¦¬λ§ κΈ°ννμμλ μμμ μ μμ μμμ λ°©ν₯μΌλ‘ μ΄λν μ μλ€. λ μ μ¬μ΄μ μΈ‘μ§μ (μ΅λ¨ κ²½λ‘)μ λ³λΆ κ³μ°λ²μΌλ‘ ꡬνλ©°, μ΄λ ν¨μ¨μ μΈ κ³μ° λ°©λ²μ΄ μ ν립λ λ¬Έμ μ΄λ€.
μ€-λ¦¬λ§ κΈ°ννμμλ λΆν¬(μ λ€λ°μ λΆλΆ λ€λ°)μ μν΄ κ²°μ λλ "μν" λ°©ν₯μΌλ‘λ§ μ΄λν μ μλ€. 물리μ μ μ½μ μμ μμ€ν
μ΄ μ νλ μ μ΄ ν΄λ°ν λμ μ§ν©μ ν΅ν΄μλ§ μ μ΄λ μ μλ€λ κ²μ΄λ€. μμμ λ°©ν₯μΌλ‘ μμ μνλ₯Ό μ§μ νμ μν¬ μ μμΌλ©°, μ€ν μ₯μΉμμ μ΄μ© κ°λ₯ν μ μ΄μ₯λ§ μΈκ°ν μ μλ€.
μ€-λ¦¬λ§ μΈ‘μ§μ λ¬Έμ λ λ¦¬λ§ μΈ‘μ§μ λ¬Έμ λ³΄λ€ κ·Όλ³Έμ μΌλ‘ μ΄λ ΅λ€:
- λͺ¨λ μ μ μ¬μ΄μ μΈ‘μ§μ μ΄ μ‘΄μ¬νμ§ μμ μ μλ€ (λ¨, Chow-Rashevskii μ 리λ κ΄νΈ μμ± λΆν¬μ λν λλ¬ κ°λ₯μ±μ 보μ₯νλ€)
- μΈ‘μ§μ μ΄ μ‘΄μ¬νλλΌλ λΉμ μ(abnormal)μΌ μ μμΌλ©°, μ΄λ λ¦¬λ§ κΈ°ννμ μ μ¬μ²΄κ° μλ λ°©μμΌλ‘ μΈ‘μ§μ λ°©μ μμ λ§μ‘±νλ€
- μΈ‘μ§μ λ°©μ μμ μ μ΄ λμμ μμ ν 리 κ΄νΈ ꡬ쑰λ₯Ό ν¬ν¨νλ―λ‘, λλΆλΆμ μμ€ν
μμ ν΄μμ μΌλ‘ λ€λ£¨κΈ° λΆκ°λ₯νλ€
λͺ¬ν
μΉ΄λ₯Όλ‘ μΈ‘μ§μ νμ
Da Silva λ±μ μ κ·Όλ²μ μ μ΄ μνμ€ κ³΅κ°μ λν λͺ¬ν
μΉ΄λ₯Όλ‘ νμμ μ¬μ©νμ¬ ν΄μμ λΆκ°λ₯μ±μ μ°ννλ€. μκ³ λ¦¬μ¦μ λ€μκ³Ό κ°λ€:
λ§€κ°λ³μν: μ μ΄ μνμ€λ₯Ό κ΅¬λΆ μμ νλ‘νμΌ(μ§μ λ μ§μ μκ° λμ μΈκ°λλ μ μ΄ μ§νμ μνμ€)λ‘ λ§€κ°λ³μννλ€
νκ°: κ° ν보μ λν΄ κ²°κ³Όμ μΈ μ λν°λ¦¬ λ³νκ³Ό μλμ§ λΉμ©μ κ³μ°νμ¬ νκ°νλ€
νμ: κ²μ΄νΈ μΆ©μ€λ(κ²°κ³Όκ° λͺ©ν κ²μ΄νΈμ μΌλ§λ κ·Όμ νμ§)μ μλμ§ μλΉ μ¬μ΄μ κ· νμ λ§μΆλ λΉμ© ν¨μμ μν΄ μ λλ λͺ¬ν
μΉ΄λ₯Όλ‘ μνλ§μΌλ‘ λ§€κ°λ³μ 곡κ°μ νμνλ€
μ μ : μ λ§ν ν보λ₯Ό κ΅μ μ΅μ ν(λ§€λλ¬μ΄ λ€μ체 SU(n) μμμμ κ²½μ¬ νκ°λ²)λ₯Ό μ¬μ©νμ¬ μ μ νλ€μ΄λ₯Ό λ€λ£¨κΈ° κ°λ₯νκ² λ§λλ κΈ°ννμ ν΅μ°°: μ μ΄ μνμ€μ μλμ§ λΉμ©μ SU(n) λ΄ λμ κ²½λ‘μ μ€-λ¦¬λ§ κΈΈμ΄μ μ§μ μ μΌλ‘ κ΄λ ¨λλ€. μλμ§ μ΅μνλ κ°μ₯ μ§§μ μ€-λ¦¬λ§ κ²½λ‘, μ¦ μΈ‘μ§μ μ μ°Ύλ κ²κ³Ό λμΉμ΄λ€. λͺ¬ν
μΉ΄λ₯Όλ‘ νμμ μ΄ μΈ‘μ§μ κ²½κ΄μ νλ₯ μ μΌλ‘ νμνμ¬, νμ€μ μΈ μμ μμ€ν
μ λν΄ ν΄μμ λ°©λ²μΌλ‘λ λλ¬ν μ μλ κ·Όμ¬ μ΅μ κ²½λ‘λ₯Ό μ°ΎμλΈλ€.
μ£Όμ₯κ³Ό κ·Όκ±°
<
| μ£Όμ₯ | κ·Όκ±° | νμ |
|---|
| μμ κ²μ΄νΈ μ΅μ νλ sub-Riemannian μΈ‘μ§μ λ¬Έμ μ΄λ€ | μνμ μ μνκ° ν립λμ΄ μλ€ | β
μ νλ¦½λ¨ |
| Monte Carlo νμμ΄ κ·Όμ΅μ μμ κ²μ΄νΈλ₯Ό μ°Ύλλ€ | Da Silva λ±μ΄ μκ·λͺ¨ μμ μμ€ν
μμ μ€μ¦νλ€ | β
μ§μ§λ¨ |
| Sub-Riemannian μΈ‘μ§μ μ Riemannian μΈ‘μ§μ κ³Ό μ§μ μΌλ‘ λ€λ₯΄λ€ | λΉμ μ μΈ‘μ§μ , bracket-generating μ μ½ μ‘°κ±΄ | β
μνμ μ¬μ€ |
| μ΄ μ κ·Όλ²μ λκ·λͺ¨ μμ μμ€ν
μΌλ‘ νμ₯λλ€ | κ³μ° λΉμ©μ΄ μμ€ν
ν¬κΈ°μ λ°λΌ μ¦κ°νλ©°, μκ·λͺ¨ μμ€ν
μμλ§ μ€μ¦λ¨ | β οΈ νμ₯μ±μ΄ κ³Όμ μ |
λ―Έν΄κ²° λ¬Έμ
νμ₯μ±: νμ¬μ μ€μ¦μ μκ·λͺ¨ μμ μμ€ν
(2~3 νλΉνΈ)μ λμμΌλ‘ νλ€. Monte Carlo μΈ‘μ§μ νμμ΄ μ€μ©μ μΈ μμ μ»΄ν¨ν
μ νμν λ€μ€ νλΉνΈ κ²μ΄νΈλ‘ νμ₯λ μ μλκ°?μ‘μ κ°κ±΄μ±: μ‘μμ΄ μλ sub-Riemannian νκ²½μμ μ°Ύμ μ΅μ κ²μ΄νΈλ μ‘μμ λ―Όκ°ν μ μλ€. κΈ°ννμ νλ μμν¬κ° μ‘μ λͺ¨λΈμ ν΅ν©νμ¬ μλμ§ μ΅μ μ±κ³Ό μ‘μ κ°κ±΄μ±μ λμμ λ§μ‘±νλ κ²μ΄νΈλ₯Ό μ°Ύμ μ μλκ°?μμμ μ₯μ λ¬Ό: SU(n)μ μμ(νΉν κΈ°λ³Έκ΅°)μ΄ νΉμ κ²μ΄νΈ μνμ€μ μ₯μ λ¬Όμ λ§λ€ μ μλ€. Sub-Riemannian ꡬ쑰λ μ΄λ¬ν μμμ μ μ½ μ‘°κ±΄κ³Ό μ΄λ»κ² μνΈμμ©νλκ°?μμ μ€λ₯ μ μ κ³Όμ μ°κ²°: μμ μ€λ₯ μ μ μ½λλ μ 체 Hilbert 곡κ°μ λΆλΆκ³΅κ°μ μ μνλ€. Sub-Riemannian μ΅μ νκ° μλμ§λ₯Ό μ΅μννλ©΄μ μ½λ κ³΅κ° λ΄μμ λμνλ κ²μ΄νΈλ₯Ό μ°Ύμ μ μλκ°?μ°κ΅¬μ λν μμ¬μ
μμ μ»΄ν¨ν
μ°κ΅¬μμκ² μμ΄, sub-Riemannian μ΅μ νλ νμ€ μ€κ³λ₯Ό μν μλ‘ μ νλ μμν¬λ₯Ό μ 곡νλ€βμ μ΄ νλΌλ―Έν°μ λν μμμ μ΅μ νλ₯Ό μμ κ²μ΄νΈμ μμ°μ λ€μ체 μμμμ κΈ°ννμ μΌλ‘ λκΈ°νλ νμμΌλ‘ λ체νλ€.
λ―ΈλΆκΈ°ννμμκ² μμ΄, μμ μ»΄ν¨ν
μ Lie κ΅° μμμ 물리μ μΌλ‘ λκΈ°νλ sub-Riemannian λ¬Έμ λ₯Ό μ 곡νλ€βκΈ°ννμ΄ νλΆνκ³ μμ©μ΄ μ¦κ°μ μΈ λ¬Έμ λ€μ λΆλ₯μ΄λ€.
References (1)
[1] da Silva, A., Castelano, L., Napolitano, R. (2025). Monte Carlo approach for finding optimally controlled quantum gates with differential geometry. Physical Review.