[pull] develop from lammps:develop

doc/src/Commands_fix.rst

@@ -101,6 +101,7 @@ OPT.
    * :doc:`indent <fix_indent>`
    * :doc:`ipi <fix_ipi>`
    * :doc:`lambda/apip <fix_lambda_apip>`
+   * :doc:`lambda/la/csp/apip <fix_lambda_la_csp_apip>`
    * :doc:`lambda_thermostat/apip <fix_lambda_thermostat_apip>`
    * :doc:`langevin (k) <fix_langevin>`
    * :doc:`langevin/drude <fix_langevin_drude>`

doc/src/Howto_apip.rst

@@ -2,7 +2,8 @@ Adaptive-precision interatomic potentials (APIP)
 ================================================
 
 The :ref:`PKG-APIP <PKG-APIP>` enables use of adaptive-precision potentials
-as described in :ref:`(Immel) <Immel2025_1>`.
+as described in :ref:`(Immel2025) <Immel2025_1>` and
+:ref:`(Immel2026) <Immel2026_2>`.
 In the context of this package, precision refers to the accuracy of an interatomic
 potential.
 
@@ -29,13 +30,13 @@ kept constant as explained below.
 
 The potential energy :math:`E_i` of an atom :math:`i` described by an
 adaptive-precision
-interatomic potential is given by :ref:`(Immel) <Immel2025_1>`
+interatomic potential is given by :ref:`(Immel2025) <Immel2025_1>`
 
 .. math::
 
    E_i = \lambda_i E_i^\text{(fast)} + (1-\lambda_i) E_i^\text{(precise)},
 
-whereas :math:`E_i^\text{(fast)}` is the potential energy of atom :math:`i`
+where :math:`E_i^\text{(fast)}` is the potential energy of atom :math:`i`
 according to a fast interatomic potential,
 :math:`E_i^\text{(precise)}` is the potential energy according to a precise
 interatomic potential and :math:`\lambda_i\in[0,1]` is the
@@ -65,14 +66,41 @@ potential is explained in :ref:`here <implementing_new_apip_styles>`.
 The switching parameter :math:`\lambda_i` that combines the two potentials
 can be dynamically calculated during a
 simulation.
+There are two ways to calculate dynamic switching parameters.
+
+1. according to :ref:`(Immel2026) <Immel2026_2>` with a differentiable
+switching parameter that results in a conservative potential.
+Energy and momentum are (in the absence of external forces) conserved
+by design.
+
+2. according to :ref:`(Immel2025) <Immel2025_1>` with a non-differentiable
+switching parameter. In this case, the implementation can be optimized for
+performance by using the switching parameters of the previous timestep.
+Thereby, one can perform most of the switching-parameter calculation within
+the force-calculation routine and include this calculations in the load-
+balancing. The potential is not conservative and energy- and
+momentum-conservation are achieved through a local correction.
+
 Alternatively, one can set a constant switching parameter before the start
 of a simulation.
+Using constant switching parameters results in a conservative potential.
+
 To run a simulation with an adaptive-precision potential, one needs the
 following components:
 
 .. tabs::
 
-   .. tab:: dynamic switching parameter
+   .. tab:: conservative dynamic switching parameter
+
+        #. :doc:`atom_style apip conservative <atom_style>` so that the switching parameter :math:`\lambda_i` can be stored.
+        #. A fast potential: :doc:`eam/apip <pair_eam_apip>` or :doc:`pace/fast/apip <pair_pace_apip>`.
+        #. A precise potential: :doc:`pace/precise/apip <pair_pace_apip>`.
+        #. :doc:`fix lambda/la/csp/apip <fix_lambda_la_csp_apip>` to calculate a differentiable switching parameter :math:`\lambda_i`.
+        #. :doc:`pair_style hybrid/overlay <pair_hybrid>` to combine the previously mentioned pair_styles.
+        #. :doc:`fix atom_weight/apip <fix_atom_weight_apip>` to approximate the load caused by every atom, as the computations of the pair_styles are only required for a subset of atoms.
+        #. :doc:`fix balance <fix_balance>` to perform dynamic load balancing with the calculated load.
+
+   .. tab:: dynamic switching parameter with correction
 
         #. :doc:`atom_style apip <atom_style>` so that the switching parameter :math:`\lambda_i` can be stored.
         #. A fast potential: :doc:`eam/apip <pair_eam_apip>` or :doc:`pace/fast/apip <pair_pace_apip>`.
@@ -102,12 +130,60 @@ Example
 .. note::
 
    How to select the values of the parameters of an adaptive-precision
-   interatomic potential is discussed in detail in :ref:`(Immel) <Immel2025_1>`.
+   interatomic potential is discussed in detail in :ref:`(Immel2025) <Immel2025_1>`
+   and :ref:`(Immel2026) <Immel2026_2>`.
 
 
 .. tabs::
 
-   .. tab:: dynamic switching parameter
+   .. tab:: conservative dynamic switching parameter
+
+      Lines like these would appear in the input script:
+
+      .. code-block:: LAMMPS
+
+         atom_style apip conservative
+         comm_style tiled
+
+         pair_style hybrid/overlay eam/fs/apip pace/precise/apip
+         pair_coeff * * eam/fs/apip Cu.eam.fs Cu
+         pair_coeff * * pace/precise/apip Cu.yace Cu
+
+         # calculate a differentiable switching parameter to obtain
+         # a conservative potential
+         fix 2 all lambda/la/csp/apip 0.24 1.5 11.0 12.0 bcc
+
+         fix 4 all atom_weight/apip 100 eam ace 0 0 all
+
+         variable myweight atom f_4
+
+         fix 5 all balance 100 1.1 rcb weight var myweight
+
+      First, the :doc:`atom_style apip conservative <atom_style>` and the communication style are set.
+
+      .. note::
+         Note, that :doc:`comm_style <comm_style>` *tiled* is required for the style *rcb* of
+         :doc:`fix balance <fix_balance>`, but not for APIP.
+         However, the flexibility offered by the balancing style *rcb*, compared to the
+         balancing style *shift*, is advantageous for APIP.
+
+      A conservative adaptive-precision EAM-ACE potential is defined based on
+      the differentiable switching parameter calculated by
+      :doc:`fix lambda/la/csp/apip <fix_lambda_la_csp_apip>`, i.e.,
+      :math:`\lambda` is calculated from the locally averaged CSP.
+      The :doc:`pair_style hybrid/overlay <pair_hybrid>` combines the fast EAM
+      and the precise ACE potential, interpolated by the corresponding
+      switching parameter.
+      Furthermore, the :doc:`fix lambda/la/csp/apip <fix_lambda_la_csp_apip>`
+      calculates the forces caused by the differentiation of the switching
+      parameter. Thereby, one obtains a conservative potential
+      (:math:`\pmb{F}_i = -\nabla_i \sum_k E_k`) that by design conserves
+      energy and momentum in the absence of external forces.
+
+      The fix 4 calculates the computational weight that is used by fix 5 to
+      balance the load between different processors.
+
+   .. tab:: dynamic switching parameter with correction
 
       Lines like these would appear in the input script:
 
@@ -222,4 +298,8 @@ of new adaptive-precision potentials.
 
 .. _Immel2025_1:
 
-**(Immel)** Immel, Drautz and Sutmann, J Chem Phys, 162, 114119 (2025)
+**(Immel2025)** Immel, Drautz and Sutmann, J Chem Phys, 162, 114119 (2025)
+
+.. _Immel2026_2:
+
+**(Immel2026)** Immel, Drautz and Sutmann, arXiv:2512.07693

doc/src/Packages_details.rst

@@ -200,6 +200,10 @@ D. Immel, R. Drautz and G. Sutmann, "Adaptive-precision potentials for
 large-scale atomistic simulations", J. Chem. Phys. 162, 114119 (2025)
 `link <immel2025_doi_>`_
 
+D. Immel, R. Drautz and G. Sutmann, "Conservative adaptive-precision
+interatomic potentials", arXiv:2512.07693
+`link <immel2026_doi_>`_
+
 Adaptive-precision means, that a fast interatomic potential, such as EAM,
 is coupled to a precise interatomic potential, such as ACE.
 This package provides the required pair_styles and fixes to run an efficient,
@@ -209,6 +213,7 @@ In the context of this package, precision refers to the accuracy of an interatom
 potential.
 
 .. _immel2025_doi: https://doi.org/10.1063/5.0245877
+.. _immel2026_doi: https://doi.org/10.48550/arXiv.2512.07693
 
 **Authors:**
 
@@ -232,6 +237,7 @@ The APIP package requires also the installation of ML-PACE, which has
 * ``examples/PACKAGES/apip``
 * :doc:`fix atom_weight/apip <fix_atom_weight_apip>`
 * :doc:`fix lambda/apip <fix_lambda_apip>`
+* :doc:`fix lambda/la/csp/apip <fix_lambda_la_csp_apip>`
 * :doc:`fix lambda_thermostat/apip <fix_lambda_thermostat_apip>`
 * :doc:`pair_style eam/apip <pair_eam_apip>`
 * :doc:`pair_style lambda/zone/apip <pair_lambda_zone_apip>`

doc/src/atom_style.rst

@@ -15,6 +15,7 @@ Syntax
   .. parsed-literal::
 
        args = none for any style except the following
+         *apip* arg = *conservative*/*thermostat* (optional) for conservative APIP/lambda thermostat APIP
          *body* args = bstyle bstyle-args
            bstyle = style of body particles
            bstyle-args = additional arguments specific to the bstyle
@@ -117,10 +118,14 @@ the Additional Information section below.
      - *bond* + "angle data"
      - :ref:`MOLECULE <PKG-MOLECULE>`
      - bead-spring polymers with stiffness
-   * - *apip*
+   * - *apip thermostat*
      - *atomic* + apip_lambda, apip_lambda_required, apip_lambda_input, apip_lambda_const, apip_lambda_input_ta, apip_e_fast, apip_e_precise, apip_f_const_lambda, apip_f_dyn_lambda
      - :ref:`APIP <PKG-APIP>`
-     - adaptive-precision interatomic potentials(APIP), see :doc:`APIP howto <Howto_apip>`
+     - adaptive-precision interatomic potentials(APIP) with a :doc:`lambda thermostat <fix_lambda_thermostat_apip>`, see :doc:`APIP howto <Howto_apip>`
+   * - *apip conservative*
+     - *atomic* + apip_lambda, apip_lambda_required, apip_e_fast, apip_e_precise
+     - :ref:`APIP <PKG-APIP>`
+     - conservative adaptive-precision interatomic potentials(APIP), see :doc:`APIP howto <Howto_apip>`
    * - *atomic*
      - tag, type, x, v, f, image, mask
      -
@@ -273,6 +278,14 @@ with both flavors of mass.
 Additional information about specific atom styles
 """""""""""""""""""""""""""""""""""""""""""""""""
 
+.. versionchanged:: TBD
+
+For the *apip* style, one can choose between the style for
+conservative potentials and the style for the
+:doc:`lambda thermostat <fix_lambda_thermostat_apip>`.
+The :doc:`Howto apip <Howto_apip>` describes the differences between
+both use cases.
+
 For the *body* style, the particles are arbitrary bodies with internal
 attributes defined by the "style" of the bodies, which is specified by
 the *bstyle* argument.  Body particles can represent complex entities,
@@ -485,6 +498,8 @@ Default
 
 The default atom style is *atomic*.  If atom_style *sphere* or
 *bpm/sphere* is used, its default argument is 0.
+If atom_style *apip* is used, its default argument is
+'thermostat'.
 
 ----------
 

doc/src/compute_property_atom.rst

@@ -35,7 +35,8 @@ Syntax
                              vfrac, s0, espin, eradius, ervel, erforce,
                              rho, drho, e, de, cv, buckling,
                              apip_lambda, apip_lambda_input, apip_e_fast,
-                             apip_e_precise
+                             apip_e_precise, apip_la_inp, apip_la_avg,
+                             apip_la_norm
 
   .. parsed-literal::
 
@@ -78,6 +79,9 @@ Syntax
            *apip_lambda* = switching parameter
            *apip_lambda_input* = input used to calculate the switching parameter
            *apip_e_fast,apip_e_precise* = potential energies mixed by the adaptive-precision potential
+           *apip_la_inp* = input used for the local averaging of the descriptor
+           *apip_la_norm* = locally averaged radial weighting function used for normalization
+           *apip_la_avg* = locally averaged descriptor used to calculate the switching parameter
 
   .. parsed-literal::
 
@@ -173,10 +177,19 @@ triangular particles and define the corner points of each triangle.
 
 The accessible quantities from the :doc:`APIP package <Howto_apip>` are
 explained in the doc pages of this package in detail.
-In short: *apip_lambda* is the switching parameter :math:`\lambda\in[0,1]`,
-that is calculated from *apip_lambda_input* and that mixes the energies
-of a fast (*apip_e_fast*) and a precise (*apip_e_precise*) potential
-into an adaptive-precision energy.
+In short: *apip_lambda* is the switching parameter :math:`\lambda\in[0,1]`.
+The switching parameter can be calculated from *apip_lambda_input* and mixes
+the energies of a
+fast (*apip_e_fast*) and a precise (*apip_e_precise*) potential into an
+adaptive-precision energy.
+
+.. versionchanged:: TBD
+
+Alternatively, the switching parameter can be calculated from a
+locally averaged descriptor (*apip_la_avg*) to obtain a conservative
+potential.
+The descriptor is calculated from an atomic property (*apip_la_inp*) and
+normalized with a locally averaged weighting function (*apip_la_norm*).
 
 .. note::
 

doc/src/fix.rst

@@ -280,6 +280,7 @@ accelerated styles exist.
 * :doc:`indent <fix_indent>` - impose force due to an indenter
 * :doc:`ipi <fix_ipi>` - enable LAMMPS to run as a client for i-PI path-integral simulations
 * :doc:`lambda/apip <fix_lambda_apip>` - compute switching parameter, that controls the precision of an :doc:`APIP potential <Howto_apip>`
+* :doc:`lambda/la/csp/apip <fix_lambda_la_csp_apip>` - compute a conservative switching parameter, that controls the precision of an :doc:`APIP potential <Howto_apip>`
 * :doc:`langevin <fix_langevin>` - Langevin temperature control
 * :doc:`langevin/drude <fix_langevin_drude>` - Langevin temperature control of Drude oscillators
 * :doc:`langevin/eff <fix_langevin_eff>` - Langevin temperature control for the electron force field model

doc/src/fix_lambda_apip.rst

@@ -198,7 +198,7 @@ Restart, fix_modify, output, run start/stop, minimize info
 The saved history of the switching parameter :math:`\lambda_i`
 and the saved history of
 :math:`\lambda_i^\text{input}` are written to
-:doc:`binary restart files <restart>` allow a smooth restart of a simulation.
+:doc:`binary restart files <restart>` to allow a smooth restart of a simulation.
 None of the :doc:`fix_modify <fix_modify>` options are relevant to this fix.
 
 If the *store_atomic_stats* argument is used, basic statistics is provided as