---
title: In-Medium Similarity Renormalization Group (IMSRG)
url: https://www.emergentmind.com/topics/in-medium-similarity-renormalization-group-imsrg
type: topic
---

# In-Medium Similarity Renormalization Group (IMSRG)

The In-Medium Similarity Renormalization Group (IMSRG) is an *ab initio* many-body framework that applies a continuous unitary transformation to a nuclear Hamiltonian, normal-ordered with respect to a chosen reference state, in order to suppress off-diagonal couplings and decouple targeted sectors of the many-body problem. In practice, the method has been used both to extract correlated ground-state energies and to derive effective Hamiltonians and operators for valence-space calculations, while later extensions have incorporated correlated references, deformation, continuum degrees of freedom, explicit normal-ordered three-body operators, infinite matter, and finite temperature [1512.06956, 1607.06882, 1805.09221].

## 1. Formal basis of the IMSRG

At its core, the IMSRG defines a flow of Hamiltonians
\[
H(s)=U(s)H(0)U^\dagger(s),
\qquad
\frac{dH(s)}{ds}=[\eta(s),H(s)],
\]
with anti-Hermitian generator
\[
\eta(s)=\frac{dU(s)}{ds}U^\dagger(s)=-\eta^\dagger(s).
\]
The flow parameter $s$ labels a continuous family of unitarily equivalent Hamiltonians. The guiding idea is to partition the Hamiltonian into “diagonal” and “off-diagonal” pieces, $H=H^{d}+H^{od}$, and choose $\eta(s)$ so that $H^{od}(s)\to 0$ as $s\to\infty$ [1512.06956, 1805.09221].

The in-medium aspect enters through normal ordering with respect to a finite-density reference state $|\Phi\rangle$, typically a Hartree–Fock Slater determinant in single-reference applications. In second quantization, the normal-ordered Hamiltonian is written as
\[
H
=
E
+\sum_{ij} f_{ij}\{a_i^\dagger a_j\}
+\frac14\sum_{ijkl}\Gamma_{ijkl}\{a_i^\dagger a_j^\dagger a_l a_k\}
+\frac1{36}\sum_{ijklmn}W_{ijklmn}\{a_i^\dagger a_j^\dagger a_k^\dagger a_n a_m a_l\},
\]
where $E$, $f$, $\Gamma$, and $W$ are the zero-, one-, two-, and three-body normal-ordered terms. Normal ordering folds substantial medium dependence, including contributions from three-nucleon forces, into low-rank operators [1512.06956, 1212.1190].

For closed-shell ground-state decoupling, the off-diagonal Hamiltonian is usually defined as the set of matrix elements that couple the reference to $1p1h$ and $2p2h$ excitations,
\[
H_{od}\equiv
\sum_{ph} f^p_h \{A^p_h\}
+\frac14\sum_{pp'hh'} \Gamma^{pp'}_{h'h}\{A^{pp'}_{hh'}\}
+\text{H.c.},
\]
so that the evolved zero-body term $E(s\to\infty)$ approximates the correlated ground-state energy [1805.09221, 1607.06882]. In valence-space formulations, the decoupling target is instead a shell-model $P$ space, and the effective Hamiltonian is the projected evolved operator,
\[
H_{\mathrm{eff}} = P\,H(\infty)\,P,
\]
which can then be diagonalized in configuration interaction [1203.2515, 1607.06882].

## 2. Truncation hierarchy, generators, and the Magnus formulation

The exact flow generates operators of arbitrarily high rank, so practical IMSRG calculations rely on systematic truncations. The standard IMSRG(2) retains normal-ordered zero-, one-, and two-body operators and discards explicit three-body terms during the flow. In conjunction with the normal-ordered two-body approximation (NO2B), this has become the workhorse formulation for medium-mass nuclei [1212.1190, 1512.06956]. Reviews of the perturbative content emphasize that IMSRG(2) reproduces the complete energy through third order in MBPT and a subset of fourth-order contributions, while remaining polynomial-scaling [1512.06956].

A full IMSRG(3) keeps the normal-ordered three-body operator $W(s)$ explicitly. The 2021 extension to IMSRG(3) established that the method is fourth-order complete in MBPT in the NO2B setting and benchmarked a hierarchy of lower-cost approximations, including IMSRG(3)-MP4, IMSRG(3)-$N^7$, IMSRG(3)-$N^8$, and IMSRG(3)-$g^5$ [2102.11172]. A later calcium study used the IMSRG(3)-$N^7$ truncation in realistic spaces and found that it yields smaller many-body uncertainties and a significantly better description of the first $2^+$ excitation energy of $^{48}$Ca than IMSRG(2), while leaving problematic calcium isotope shifts largely unresolved [2411.16014].

Generator choice governs both numerical behavior and the character of the decoupling. The Wegner generator,
\[
\eta=[H_d,H_{od}],
\]
guarantees monotonic suppression of off-diagonal couplings but can produce stiff differential equations [1512.06956]. White-type generators use ratios of off-diagonal matrix elements to energy denominators and typically yield faster, less stiff flows. Imaginary-time generators replace the denominator by $\mathrm{sgn}(\Delta)$, while Brillouin generators are particularly natural for correlated multireference states because they avoid an explicit diagonal/off-diagonal partition [1408.6555, 1805.09221]. In deformed single-reference IMSRG, the White generator is used together with the off-diagonal set $\{f_{ph},\Gamma_{pp'hh'}\}$ in the deformed basis [2204.07301].

The Magnus formulation rewrites the unitary transformation as
\[
U(s)=e^{\Omega(s)},
\qquad
\Omega^\dagger=-\Omega,
\]
so that
\[
\tilde H=e^\Omega H e^{-\Omega}
=
H+[\Omega,H]+\frac1{2!}[\Omega,[\Omega,H]]+\dots
\]
and similarly for any operator $O$. This reformulation reduces memory requirements, permits low-order ODE integrators without loss of accuracy in the transformed observables, and makes consistent operator evolution essentially routine once $\Omega(s)$ has been obtained [1507.06725, 2204.07301].

## 3. Reference states, multireference formulations, and symmetry restoration

The simplest IMSRG implementation uses a single Slater determinant reference and is therefore most natural for closed-shell nuclei. Open-shell and collective systems require more elaborate references. The multireference IMSRG (MR-IMSRG) generalizes normal ordering and Wick’s theorem to correlated reference states by introducing irreducible density matrices $\lambda^{(n)}$ and has been used with particle-number-projected Hartree–Fock–Bogoliubov references to describe even calcium and nickel isotopes with chiral NN+3N interactions [1408.6555, 1607.06882].

The generalized formalism expands the Hamiltonian as
\[
\hat H = E + \sum_{pq} f^p_q \{A^p_q\}
+ \frac14 \sum_{pqrs}\Gamma^{pq}_{rs}\{A^{pq}_{rs}\}+\cdots,
\]
with cumulants such as
\[
\lambda^p_q=\rho^p_q,
\qquad
\lambda^{pq}_{rs}=\rho^{pq}_{rs}-\mathcal A(\lambda^p_r\lambda^q_s),
\]
entering both the flow equations and the Brillouin generator [1807.11053, 1408.6555]. In MR-IMSRG(2), terms nonlinear in $\lambda^{(2)}$ and those involving $\lambda^{(3)}$ may be dropped depending on the reference and implementation, whereas in IMSRG+GCM calculations inclusion of $\lambda^{(3)}$ in the Brillouin generator is necessary for convergence in $^{48}$Ti [1807.11053].

A distinct strategy for open-shell nuclei is to break rotational symmetry at the reference level. The deformed IMSRG (D-IMSRG) uses an axially deformed Hartree–Fock Slater determinant in an $m$-scheme basis, thereby constructing a single-reference IMSRG for even-even open-shell systems. The corresponding reference preserves axial, parity, and time-reversal symmetries while breaking spherical rotational symmetry. Static rotational correlations are then approximated by angular-momentum projection of the HF state,
\[
P^J_{MM'}=\frac{2J+1}{8\pi^2}\int d\omega\, D^{J*}_{MM'}(\omega)R(\omega),
\]
with leading-order correction
\[
\Delta E_{\rm proj}
=
\frac{\langle\Phi|HP|\Phi\rangle}{\langle\Phi|P|\Phi\rangle}
-
\frac{\langle\Phi|H|\Phi\rangle}{\langle\Phi|\Phi\rangle}.
\]
This avoids the much more expensive full projection of the correlated IMSRG state [2204.07301].

Another extension uses symmetry-restored generator-coordinate-method references. IMSRG+GCM starts from deformed HFB states projected on good angular momentum and particle number,
\[
|\Psi^{J^\pi}_{N,Z}\rangle
=
\sum_q f_q\,\hat P^J_{MK}\hat P^N\hat P^Z|\Phi(q)\rangle,
\]
and then applies MR-IMSRG(2) with a Brillouin generator. In pf-shell benchmarks this improves low-lying spectra relative to either spherical-reference IMSRG or GCM with unevolved operators, and it provides a consistent route to evolved $0\nu\beta\beta$ operators via the PI and PF prescriptions [1807.11053].

| Variant | Reference state | Primary target |
|---|---|---|
| Single-reference IMSRG | HF Slater determinant | Closed-shell ground states |
| MR-IMSRG | Projected HFB or correlated reference | Open-shell ground states |
| VS-IMSRG | Core reference plus valence-space decoupling | Shell-model Hamiltonians |
| D-IMSRG | Axially deformed HF Slater determinant | Deformed even-even nuclei |
| IMSRG+GCM | Symmetry-restored GCM states | Collective spectra and $0\nu\beta\beta$ |

## 4. Evolved operators, excited states, and transition observables

A central technical advantage of the IMSRG is that observables can be evolved consistently with the Hamiltonian. In Magnus form,
\[
\tilde O
=
e^\Omega O e^{-\Omega}
=
O+[\Omega,O]+\frac1{2!}[\Omega,[\Omega,O]]+\dots,
\]
and matrix elements are evaluated in the decoupled reference or in post-IMSRG many-body states [2204.07301, 1705.05511]. This is essential for charge radii, moments, transition strengths, and weak matrix elements.

For charge radii, D-IMSRG uses the intrinsic decomposition
\[
\langle R_{\rm ch}^2\rangle
=
\langle R_p^2\rangle
+\langle r_p^2\rangle
+\frac{N}{Z}\langle r_n^2\rangle
+\langle r_{\rm DF}^2\rangle
+\langle r_{\rm so}^2\rangle,
\]
with $\langle r_p^2\rangle=0.709\ \mathrm{fm}^2$, $\langle r_n^2\rangle=-0.106\ \mathrm{fm}^2$, and $\langle r_{\rm DF}^2\rangle=0.033\ \mathrm{fm}^2$ in that implementation [2204.07301]. A more recent calcium study used
\[
R_{\rm ch}^2=\langle R_p^2\rangle+\langle r_{\rm so}^2\rangle+r_p^2+\frac{N}{Z}r_n^2+\frac{3}{4M^2},
\]
together with a corrected spin-orbit term proportional to $\mu_i-Q_i/2$ [2411.16014].

Excited states can be accessed either by deriving effective interactions for subsequent configuration-interaction diagonalization or by working directly with the decoupled Hamiltonian. EOM-IMSRG formulates
\[
[\bar H,\bar X^\dagger_\nu(J^\pi)]|\Phi_0\rangle
=
(E_\nu-E_0)\bar X^\dagger_\nu(J^\pi)|\Phi_0\rangle,
\]
with a ladder operator truncated, for example, at the $1p1h+2p2h$ level. TDA-IMSRG and valence-restricted TDA-IMSRG instead engineer additional decouplings so that a Tamm–Dancoff diagonalization becomes exact in a chosen sector, though EOM-IMSRG is more robust numerically [1611.00661].

For electromagnetic observables, the consistent IMSRG evolution of tensor operators was implemented in both EOM-IMSRG(2,2) and VS-IMSRG(2). Magnetic dipole observables are generally in reasonable agreement with experiment, whereas the more collective electric quadrupole and octupole observables are significantly underpredicted, often by over an order of magnitude, indicating missing physics at the present truncation level [1705.05511]. In the pf-shell benchmark for neutrinoless double-beta decay, IMSRG+GCM gives better spectra than either spherical-reference IMSRG or GCM with unevolved operators, while the improvement in the $0\nu\beta\beta$ matrix element is slight in the single-shell setup; the dominant collective isoscalar-pairing correlations must be included explicitly in the GCM coordinates [1807.11053].

## 5. Finite nuclei: closed shells, valence spaces, deformation, and continuum

Applications to finite nuclei span closed-shell energies, open-shell spectroscopy, deformation, and open quantum systems. Early chiral NN+3N calculations for closed shells up to $^{56}$Ni showed good agreement with experiment in $^4$He and the closed-shell oxygen isotopes, while calcium and nickel isotopes were somewhat overbound; comparison with coupled-cluster and importance-truncated no-core shell model established the accuracy of IMSRG(2) and the role of the NO2B approximation [1212.1190]. MR-IMSRG calculations across even calcium and nickel isotopes subsequently demonstrated the importance of chiral 3N interactions for correct shell evolution and dripline trends, with agreement at shell closures comparable to CC results obtained with the same Hamiltonians [1408.6555].

The valence-space IMSRG provides a nonperturbative derivation of shell-model Hamiltonians. The original open-shell application derived effective $p$- and $sd$-shell Hamiltonians for $^6$Li and $^{18}$O, obtaining a spectrum for $^6$Li in very good agreement with *ab initio* results and establishing a direct alternative to diagrammatic effective-interaction theory for shell-model studies [1203.2515]. Later reviews emphasized successful spectroscopy in the lower $sd$ shell and access to intrinsically deformed states through shell-model diagonalization [1607.06882].

For deformed nuclei, D-IMSRG calculations with NNLO$_{\rm opt}$ showed that projected HF corrections lower the energies of $^{8\text{–}16}$Be by about $5$–$6$ MeV, with specific corrections
\[
\Delta E_{\rm proj}(^{8}\mathrm{Be})=-7.2\ \mathrm{MeV},
\qquad
\Delta E_{\rm proj}(^{10}\mathrm{Be})=-5.9\ \mathrm{MeV},
\]
and gains of about $3$–$6$ MeV in Ne and Mg near the $N=20$ island of inversion. In that study, both D-IMSRG and VS-IMSRG predicted the even-even Be dripline at $^{12}$Be, whereas experiment places it at $^{14}$Be; the omission of continuum coupling was explicitly identified as a likely source of the discrepancy [2204.07301].

Continuum degrees of freedom can be incorporated directly through the Berggren basis. The Gamow IMSRG reformulates the flow for a complex-symmetric Hamiltonian, using a complex-orthogonal similarity transformation in a basis that satisfies the Berggren completeness relation. In $^{24}$O it reproduces the observed resonant structure, including the triplet of $2^+$, $3^+$, and $4^+$ states around $E_x\approx 8$ MeV. In $^{22}$C it predicts a pronounced halo density and demonstrates that the neutron $s_{1/2}$ continuum is essential: the matter radius increases from $2.798$ to $2.928$ fm with NNLO$_{\rm opt}$ and from $2.983$ to $3.139$ fm with NNLO$_{\rm sat}$ when the $s$-wave is promoted from discrete states to a Berggren continuum [1906.10539].

Recent IMSRG(3)-$N^7$ calculations for $^{44,48,52}$Ca sharpened the method’s quantitative status. For $^{48}$Ca, the first $2^+$ excitation in VS-IMSRG(2) is $4.930$ MeV, and the IMSRG(3)-$N^7$ correction is $-0.677$ MeV, moving the result toward the experimental $3.832$ MeV and improving the description of the $N=28$ shell closure. By contrast, charge-radius corrections are small: for $^{48}$Ca the single-reference radius changes from $3.300$ fm by only $+0.018$ fm, and the large experimental increase in $R_{\rm ch}(^{52}\mathrm{Ca})-R_{\rm ch}(^{48}\mathrm{Ca})$ remains unexplained [2411.16014].

## 6. Infinite matter and finite-temperature extensions

The IMSRG has also been extended beyond finite nuclei. A 2025 development formulated IMSRG(2) directly in homogeneous infinite nuclear matter using a plane-wave basis around a filled Fermi sea, normal ordering at finite density, and a White generator with Epstein–Nesbet denominators. The reference occupations are those of the Fermi sphere, and the Hamiltonian is written as
\[
H(s)=E(s)+\sum_{ij} f_{ij}(s)\,a_i^\dagger a_j
+\frac14\sum_{ijkl}\Gamma_{ijkl}(s)\,a_i^\dagger a_j^\dagger a_l a_k+\cdots .
\]
For pure neutron matter with the soft NNLO$_{\rm opt}$ interaction, IMSRG(2) and CCD agree very well up to $\rho\lesssim 0.28\ \mathrm{fm}^{-3}$ for NN only and up to $\rho\lesssim 0.30\ \mathrm{fm}^{-3}$ when 3N forces are included through NO2B. For a harder N$^3$LO interaction with $\Lambda=700$ MeV, pronounced discrepancies emerge among MBPT3, MBPT4, and IMSRG(2), especially above $\rho\gtrsim 0.2\ \mathrm{fm}^{-3}$, while in symmetric nuclear matter even soft interactions already show large differences among MBPT, CCD, and IMSRG(2) near and above saturation density [2501.02479].

A finite-temperature extension generalizes the reference from a zero-temperature Slater determinant to a thermal Hartree–Fock ensemble, with occupations
\[
n_i=\frac{1}{1+e^{\beta(\epsilon_i-\mu)}}
\]
in the grand canonical ensemble. The IMSRG(2) flow equations retain their form, but the integer occupations are replaced by thermal factors. Because a straightforward White generator can become ill-behaved when occupations appear in both numerators and denominators, the finite-temperature implementation adopts an arctan-regularized White generator with occupation-dependent denominators and preserves the IMSRG flow equation
\[
\frac{dH}{ds}=[\eta,H]
\]
in the thermal setting [2407.00576].

Benchmarking on an exactly solvable pairing-plus-particle–hole model showed that FT-IMSRG substantially improves over FT-HF across temperatures, with relative errors typically below the percent level for weak to moderate couplings and the largest deviations at intermediate $\beta\sim 1$. Canonical and grand canonical implementations differ more strongly in small systems, but those differences diminish rapidly with particle number, which suggests that the grand canonical ensemble is the practical choice for realistic nuclear applications [2407.00576].

## 7. Computational strategies, uncertainty quantification, and outlook

Because IMSRG calculations are dominated by tensor contractions in the two- and three-body sectors, algorithmic developments have become a major subfield. Importance truncation for IMSRG(2) selects only “important” two-body matrix elements according to a measure $\kappa_{pqrs}$ and leaves the rest out of the flow. The most effective strategy in the benchmark study was to truncate only the $hppp$ and $pppp$ blocks, using either
\[
\kappa^{\Gamma/\varepsilon}_{pqrs} = |\Gamma_{pqrs}/\varepsilon_{\rm sum}|
\]
or
\[
\kappa^{\Gamma n}_{pqrs}=|\Gamma_{pqrs}| \times \kappa^n_{pqrs},
\]
together with a perturbative third-order correction $E^{(3)}_{\rm IT}$ as an error indicator. For soft Hamiltonians and large single-particle spaces, compression ratios up to $R_C\approx 100$ are feasible with sub-MeV discrepancies, while hard interactions produce larger and more conservative error estimates [2110.09390].

Machine-learning acceleration has also been explored. IMSRG-Net learns the Magnus operator $\Omega(s)$ directly from a short training window of the flow, using a loss function that combines data fidelity for $\Omega$ and its derivative-based generator proxy. In benchmarks for $^{16}$O and $^{40}$Ca, training on ten points up to $s=20$ sufficed to reproduce full IMSRG(2) ground-state energies with typical absolute differences below $1$ keV and charge radii with typical absolute differences of order $10^{-4}$ fm. The same strategy was successfully applied to the second step of VS-IMSRG decoupling [2306.08878].

The numerical treatment of the Magnus expansion itself remains nontrivial. A 2026 analysis of the hunter-gatherer scheme showed that, although it reduces storage to two Magnus operators, it can differ from standard IMSRG(2) approaches by up to $7\,\mathrm{MeV}$ for ground-state energies and $0.5\,\mathrm{MeV}$ for excitation energies, with the discrepancies in some cases comparable to expected IMSRG(3) corrections. The paper therefore recommends split Magnus or direct-flow schemes when the target accuracy is at the level of IMSRG(3) improvements, especially in valence-space applications with large decouplings [2601.16133].

Uncertainty quantification is now increasingly explicit. In calcium, IMSRG(3)-$N^7$ corrections imply practical IMSRG(2) uncertainty estimates of approximately $2$–$3\%$ for correlation energies, $1$–$1.5\%$ for charge radii, and $5$–$7.5\%$ for neutron skins for soft interactions such as 1.8/2.0 (EM) [2411.16014]. Earlier MR-IMSRG studies estimated beyond-(2) effects in two-neutron separation energies at the few-hundred-keV level by comparison to CCSD and CR-CC(2,3), while NO2B overbinding in calcium and nickel was found to be below $1\%$ in total energies [1408.6555]. These benchmarks suggest that many-body truncation, omitted residual and induced three-body terms, continuum coupling, and deficiencies of current chiral Hamiltonians contribute at comparable levels depending on the observable.

The emerging picture is that the IMSRG is no longer a single approximation but a family of controlled transformations. Single-reference, multireference, valence-space, deformed, Gamow, finite-temperature, and infinite-matter variants all preserve the same central idea—continuous in-medium decoupling—while differing in reference choice, generator, and operator-rank truncation. Current developments point toward fuller symmetry restoration, more systematic treatment of explicit and induced three-body operators, broader continuum embeddings, and tighter integration of uncertainty quantification with the evolving hierarchy from IMSRG(2) to IMSRG(3) and beyond [1805.09221, 2102.11172, 2411.16014].

Source: https://www.emergentmind.com/topics/in-medium-similarity-renormalization-group-imsrg