Kenneth G. Wilson's 1975 survey 'Renormalization group methods' posed a program: turn the renormalization group (RG) — the successive elimination of short-scale degrees of freedom of a multiscale problem, with their effects folded into effective couplings — into a practical numerical method. At the time, his numerical renormalization group (NRG) had succeeded on only two problems, the Kondo problem and the two-dimensional Ising model, while turbulence, the molecular bond of large molecules, and essentially all of quantum field theory had no usable numerical RG; even the three-dimensional Ising model required summing over an astronomical number of spin configurations per RG step. Wilson asked for (i) RG formulations whose retained degrees of freedom per iteration stay manageable, possibly by combining RG with Monte Carlo sampling, and (ii) a systematic, non-artistic procedure for constructing RG transformations. The named instances have since been largely resolved by the subsequent lineage — Monte Carlo RG, the density-matrix RG (standard for one-dimensional quantum systems and competitive ab initio quantum chemistry), the generalized NRG for quantum impurity problems, tensor-network renormalization, and the functional RG — leaving a precise open core: there is still no practical, controlled numerical RG that reduces developed turbulence to the point where numerical methods can solve it; nor for strongly coupled field theories in the regimes lattice Monte Carlo cannot reach (finite baryon density, real time), where the fermion sign problem is NP-hard; nor, in general, a systematic non-artistic procedure for constructing RG transformations for new problems with an account of fixed-point convergence, worst-case uncomputability of RG flows ruling out a fully general cookbook; and the molecular-bond ask survives to the extent that bond-level effective interactions for large molecules remain beyond existing ab initio quantum chemistry. The problem is open.

Background

Many physical problems couple widely separated length or energy scales. Near a continuous phase transition the correlation length (the typical size of ordered regions) diverges, so all scales contribute simultaneously; in turbulent flow, vortices exist at every size between the driving scale and the viscous dissipation scale, and its statistical description involves the energy spectrum, dissipation-range observables, and intermittent multi-point statistics (fluctuations of small-scale quantities that deviate from self-similar scaling); around a magnetic impurity in a metal, the screening cloud of conduction electrons extends across decades of energy; and in quantum field theory the original 'renormalization' problem — divergent loop integrals — is itself a short-versus-long-scale conflict. The renormalization group (RG) exploits this structure: instead of treating all scales at once, one repeatedly eliminates the shortest-scale degrees of freedom, folding their effect into effective interactions (couplings) for the remaining ones, and studies the induced flow of the couplings as the resolution scale is lowered. A fixed point of this flow, a coupling set that maps to itself, governs universal behavior such as critical exponents.

In the early 1970s Wilson turned the RG into a numerical method. For the Kondo problem — a single magnetic impurity exchange-coupled to the conduction electrons of a nonmagnetic metal, whose thermodynamics had resisted perturbation theory because the effective coupling grows logarithmically at low temperature — Wilson discretized the conduction band into logarithmically scaled energy shells and iteratively diagonalized the resulting sequence of Hamiltonians, keeping only the lowest-lying many-body states at each step. This numerical renormalization group (NRG) solved the problem and exposed the low-temperature behavior directly. Separately, Niemeijer and van Leeuwen (1973) had implemented Wilson's ideas numerically for the two-dimensional triangular-lattice Ising model (spins taking values ±1\pm 1 with nearest-neighbor coupling) via a block-spin transformation: grouping three spins into one effective spin, summing over the block's internal configurations to obtain renormalized couplings, and iterating toward a fixed point whose eigenvalues yield critical exponents.

The obstacle to generalizing either construction is combinatorial. A block-spin transformation on a block of nn spins requires summing over 2n2^{n} internal configurations, and each RG step generates new couplings (multi-spin products, longer-range terms), so the coupling space grows with every iteration; truncating it was uncontrolled, and there was no principle guaranteeing the flow reaches a fixed point at all. For two dimensions with three-spin blocks this bookkeeping was feasible; Wilson's companion review of 1975 noted that the three-dimensional analogue 'would contain over 30 spins', for which 'the number of configurations needed becomes astronomical'.

Wilson's 1975 survey 'Renormalization group methods' took stock of this situation. Numerical RG had succeeded on exactly two problems (Kondo, two-dimensional Ising), while the multiscale problems that most needed it — turbulence, the molecular bond of large molecules (quantum chemistry, where conventional ab initio (first-principles) electronic-structure computations scale badly with system size), and, in his words, 'all of quantum field theory' — had no practical numerical RG. He wrote that the difficulties of renormalization prevent one from formulating 'even a Monte Carlo method for quantum electrodynamics' (QED, the quantum field theory of electromagnetism), that the RG was 'at present an approach of last resort', and that two things were missing: first, RG formulations in which the number of degrees of freedom retained per iteration stays manageable — possibly by combining RG with Monte Carlo sampling (stochastic sampling of configuration space, which sidesteps explicit summation over all configurations); and second, a systematic procedure for constructing RG transformations, since their invention was then an art with no cookbook.

The methods that later grew out of this program are: Monte Carlo RG (MCRG), which applies the RG transformation to Monte Carlo-generated configurations so that blocked couplings are measured rather than summed exactly; the density-matrix renormalization group (DMRG), which improves Wilson's state selection for quantum lattice problems by retaining the eigenstates of the reduced density matrix of the system-environment split — the states most entangled with the kept block; the numerical renormalization group in its modern generalized form for quantum impurity models; tensor-network renormalization (TNR), which coarse-grains tensor networks (partition functions or quantum states written as arrays of tensors with contracted indices) in a way that preserves the fixed-point structure at criticality; and the functional or exact renormalization group (FRG), an exact flow equation for a scale-dependent effective action from which nonperturbative approximations can be built systematically. Two obstructions discovered later frame the surviving core: the fermion sign problem — Monte Carlo sampling weights that become non-positive for many fermion systems at nonzero chemical potential (the grand-canonical parameter controlling particle density), making the sampling cost grow exponentially — which was proved NP-hard; and results from the undecidability of the spectral gap (the energy difference between a quantum spin system's ground state and first excited state) implying that worst-case RG flows can be uncomputable and that no fully general recipe for constructing renormalization-group transformations can be guaranteed.

Problem Statement

The problem is Wilson's 1975 program: make the renormalization group into a practical numerical method for the multiscale problems it was created for. It has three linked asks. (1) Construct explicit renormalization-group transformations whose retained degrees of freedom per iteration remain bounded and computably tractable as system size grows — unlike the 1975 real-space block-spin scheme, where a three-dimensional Ising calculation would already require dealing with blocks of more than 30 spins and an astronomical number of internal configurations — with controlled truncation error, for at least one problem class that Wilson named as then lacking any practical numerical RG: developed turbulence, the electronic structure of large molecules at the bond level, or strongly coupled quantum field theory (field theory whose interaction strength admits no small parameter, so that perturbation expansions fail). (2) Formulate the renormalization-group step so that it remains numerically implementable where direct summation over all configurations is impossible. (3) Replace the case-by-case artistic invention of renormalization-group transformations with a systematic construction procedure, with an account of when the constructed transformations converge to fixed points. Later literature resolved asks (1)-(2) for equilibrium classical critical phenomena, quantum impurity models, one-dimensional quantum lattice systems, ab initio quantum chemistry for the molecular systems the density-matrix renormalization group already treats, two-dimensional classical critical systems, and, as a framework, nonperturbative field theories via the functional RG. The surviving open core of the problem is: (a) a practical, controlled numerical RG for developed turbulence (turbulent flow with a wide separation between the large scale of energy input and the small viscous dissipation scale) that reduces the problem's complexity — the number of degrees of freedom that must be numerically simulated — to the point where numerical methods can be used to solve it, with controlled truncation error; (b) a practical, controlled numerical RG for strongly coupled quantum field theories in at least one of the regimes where lattice Monte Carlo (sampling of field configurations on a discretized spacetime lattice) fails, namely nonzero baryon chemical potential (the grand-canonical parameter controlling the net density of baryons, i.e., protons and neutrons) or real-time dynamics (lattice Monte Carlo is formulated in imaginary time, from which real-time evolution cannot be directly recovered); (c) a systematic, non-artistic procedure for constructing renormalization-group transformations for new problems, together with an account of when the constructed transformations converge to fixed points; and (d) the molecular-bond ask in the source's sense, for large molecules whose bond-level effective interactions remain beyond existing ab initio quantum-chemistry methods: a practical, controlled numerical RG delivering an effective interaction at the bond level for such molecules. For the numerical-RG parts (a), (b), and (d), the same conditions of practicality and control hold as for the resolved instances — bounded cost per iteration, controlled error, and, where independent reference data exist, quantitative agreement with them — concretely, a retained-degree-of-freedom count per iteration that stays bounded as system size grows, together with a stated truncation-error bound, verified against independent reference data where such data exist. The construction part (c) requires a construction procedure that yields, for at least one new problem for which no practical, controlled numerical renormalization group previously existed (one of the problem classes in (a), (b), or (d), or a problem of comparable difficulty), a renormalization-group transformation with the retained-degree-of-freedom and truncation-error guarantees required of (a), (b), and (d), together with an account of when the constructed transformations converge to fixed points.

References

  1. K. G. Wilson, 'Renormalization group methods', Advances in Mathematics 16, 170-186 (1975). DOI: 10.1016/0001-8708(75)90149-8, https://doi.org/10.1016/0001-8708(75)90149-8
  2. K. G. Wilson, 'The renormalization group: Critical phenomena and the Kondo problem', Reviews of Modern Physics 47, 773-840 (1975). DOI: 10.1103/RevModPhys.47.773, https://doi.org/10.1103/RevModPhys.47.773
  3. K. G. Wilson, 'The renormalization group and critical phenomena', Reviews of Modern Physics 55, 583-600 (1983), published version of the Nobel Lecture of 8 December 1982. DOI: 10.1103/RevModPhys.55.583, https://doi.org/10.1103/RevModPhys.55.583
  4. Th. Niemeijer and J. M. J. van Leeuwen, 'Wilson Theory for Spin Systems on a Triangular Lattice', Physical Review Letters 31, 1411-1414 (1973). DOI: 10.1103/PhysRevLett.31.1411, https://doi.org/10.1103/PhysRevLett.31.1411
  5. Shang-keng Ma, 'Renormalization Group by Monte Carlo Methods', Physical Review Letters 37, 461-464 (1976). DOI: 10.1103/PhysRevLett.37.461, https://doi.org/10.1103/PhysRevLett.37.461
  6. Robert H. Swendsen, 'Monte Carlo Renormalization Group', Physical Review Letters 42, 859-861 (1979). DOI: 10.1103/PhysRevLett.42.859, https://doi.org/10.1103/PhysRevLett.42.859
  7. Steven R. White, 'Density Matrix Formulation for Quantum Renormalization Groups', Physical Review Letters 69, 2863-2866 (1992). DOI: 10.1103/PhysRevLett.69.2863, https://doi.org/10.1103/PhysRevLett.69.2863
  8. Steven R. White and Richard L. Martin, 'Ab initio quantum chemistry using the density matrix renormalization group', The Journal of Chemical Physics 110, 4127-4130 (1999). DOI: 10.1063/1.478295, https://doi.org/10.1063/1.478295
  9. Ulrich Schollwöck, 'The density-matrix renormalization group', Reviews of Modern Physics 77, 259-315 (2005). arXiv:cond-mat/0409292, https://arxiv.org/abs/cond-mat/0409292, DOI: 10.1103/RevModPhys.77.259, https://doi.org/10.1103/RevModPhys.77.259
  10. Ralf Bulla, Theo A. Costi, and Thomas Pruschke, 'Numerical renormalization group method for quantum impurity systems', Reviews of Modern Physics 80, 395-450 (2008). DOI: 10.1103/RevModPhys.80.395, https://doi.org/10.1103/RevModPhys.80.395
  11. G. Evenbly and G. Vidal, 'Tensor Network Renormalization', Physical Review Letters 115, 180405 (2015). DOI: 10.1103/PhysRevLett.115.180405, https://doi.org/10.1103/PhysRevLett.115.180405
  12. C. Wetterich, 'Exact evolution equation for the effective potential', Physics Letters B 301, 90-94 (1993). DOI: 10.1016/0370-2693(93)90726-X, https://doi.org/10.1016/0370-2693(93)90726-X
  13. J. Berges, N. Tetradis, and C. Wetterich, 'Non-perturbative renormalization flow in quantum field theory and statistical physics', Physics Reports 363, 223-386 (2002). arXiv:hep-ph/0005122, https://arxiv.org/abs/hep-ph/0005122, DOI: 10.1016/S0370-1573(01)00098-9, https://doi.org/10.1016/S0370-1573(01)00098-9
  14. A. Pelissetto and E. Vicari, 'Critical phenomena and renormalization-group theory', Physics Reports 368, 549-727 (2002). arXiv:cond-mat/0012164, https://arxiv.org/abs/cond-mat/0012164, DOI: 10.1016/S0370-1573(02)00219-3, https://doi.org/10.1016/S0370-1573(02)00219-3
  15. David McComb, Taek-Jin Yang, Alistair Young, and Luc Machiels, 'Investigation of renormalization group methods for the numerical simulation of isotropic turbulence', preprint (1998). arXiv:physics/9806028, https://arxiv.org/abs/physics/9806028
  16. Matthias Troyer and Uwe-Jens Wiese, 'Computational Complexity and Fundamental Limitations to Fermionic Quantum Monte Carlo Simulations', Physical Review Letters 94, 170201 (2005). DOI: 10.1103/PhysRevLett.94.170201, https://doi.org/10.1103/PhysRevLett.94.170201
  17. Léonie Canet, 'Functional renormalisation group for turbulence', Journal of Fluid Mechanics 950, P1 (2022). arXiv:2205.01427, https://arxiv.org/abs/2205.01427, DOI: 10.1017/jfm.2022.808, https://doi.org/10.1017/jfm.2022.808
  18. James D. Watson, Emilio Onorati, and Toby S. Cubitt, 'Uncomputably complex renormalisation group flows', Nature Communications 13, 7618 (2022). arXiv:2102.05145, https://arxiv.org/abs/2102.05145, DOI: 10.1038/s41467-022-35179-4, https://doi.org/10.1038/s41467-022-35179-4