Skip to content

Latest commit

 

History

History
996 lines (895 loc) · 40 KB

librarylink.rst

File metadata and controls

996 lines (895 loc) · 40 KB

Mathematica interface

The FlexibleSUSY-generated spectrum generators can be called from within Mathematica using Wolfram's LibraryLink interface.

The following example calculates the pole mass spectrum and the observables in the CMSSM for a given parameter point:

Get["models/CMSSM/CMSSM_librarylink.m"];

(* Create a handle to a model given the input parameters.
   See Options[FSCMSSMOpenHandle] for all default options. *)
handle = FSCMSSMOpenHandle[
  fsSettings -> { precisionGoal -> 1.*^-4 },
  fsSMParameters -> { Mt -> 173.3 },
  fsModelParameters -> {
      m0 -> 125, m12 -> 500, TanBeta -> 10, SignMu -> 1, Azero -> 0 }
];

(* calculate pole mass spectrum *)
FSCMSSMCalculateSpectrum[handle]

(* calculate further observables *)
FSCMSSMCalculateObservables[handle]

(* close the model handle *)
FSCMSSMCloseHandle[handle];

Note: For each model, FlexibleSUSY creates an example Mathematica script which illustrates the use of the Mathematica interface. The generated example can be found in models/<model>/run_<model>.m, which can be run for example as

math -run "<< \"models/<model>/run_<model>.m\""

Before running the example Mathematica script, the model parameters in the script should be set to reasonable values.

In the following more advanced example the Higgs pole mass is calculated in the HSSUSY model (an EFT calculation of the SM-like Higgs mass, assuming that the high-energy completion is the MSSM) as a function of X_t / M_S for \tan\beta = 5 and different values of the SUSY scale. The example also illustrates how parallelization can be used to exploit the performance of multi-core systems.

Get["models/HSSUSY/HSSUSY_librarylink.m"];

CalcMh[TB_, Xtt_, MS_] := Module[{handle, spec},
    handle = FSHSSUSYOpenHandle[
        fsSettings -> {
            precisionGoal -> 1.*^-5,
            calculateStandardModelMasses -> 1,
            poleMassLoopOrder -> 4,
            ewsbLoopOrder -> 4,
            betaFunctionLoopOrder -> 4,
            thresholdCorrectionsLoopOrder -> 3,
            thresholdCorrections -> 123111321,
            parameterOutputScale -> 173.34
        },
        fsModelParameters -> {
            TanBeta -> TB,
            MEWSB -> 173.34,
            MSUSY -> MS,
            M1Input -> MS,
            M2Input -> MS,
            M3Input -> MS,
            MuInput -> MS,
            mAInput -> MS,
            AtInput -> (Xtt + 1/TB) * MS,
            AbInput -> 0,
            AtauInput -> 0,
            msq2 -> MS^2 IdentityMatrix[3],
            msu2 -> MS^2 IdentityMatrix[3],
            msd2 -> MS^2 IdentityMatrix[3],
            msl2 -> MS^2 IdentityMatrix[3],
            mse2 -> MS^2 IdentityMatrix[3],
            LambdaLoopOrder -> 2,
            TwoLoopAtAs -> 1,
            TwoLoopAbAs -> 1,
            TwoLoopAtAb -> 1,
            TwoLoopAtauAtau -> 1,
            TwoLoopAtAt -> 1
        }
    ];
    spec = HSSUSY /. FSHSSUSYCalculateSpectrum[handle];
    FSHSSUSYCloseHandle[handle];
    If[spec =!= $Failed, Pole[M[hh]] /. spec, 0]
];

LaunchKernels[];
DistributeDefinitions[CalcMh];

data = {
    ParallelMap[{#, CalcMh[5, #, 1000 ]}&, Range[-3.5, 3.5, 0.1]],
    ParallelMap[{#, CalcMh[5, #, 2000 ]}&, Range[-3.5, 3.5, 0.1]],
    ParallelMap[{#, CalcMh[5, #, 10000]}&, Range[-3.5, 3.5, 0.1]]
};

plot = ListPlot[data,
                PlotLegends -> {"MS = 1 TeV", "MS = 2 TeV", "MS = 10 TeV"},
                Axes -> False, Frame -> True,
                FrameLabel -> {"Xt / MS", "Mh / GeV"}];

Export["HSSUSY_Mh_Xt.png", plot, ImageSize -> 1000];

Output:

images/HSSUSY_Mh_Xt.png

In order to build the LibraryLink library, FlexibleSUSY must be configured with --enable-meta (enabled by default).

Example:

./configure --with-models=CMSSM
make

The LibraryLink library can be found in models/<model>/<model>_librarylink.so, where <model> is the model name. In order to use FlexibleSUSY's generated <model> spectrum generator at the Mathematica level, the library functions must be loaded using the models/<model>/<model>_librarylink.m script.

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];

First, a handle to the model must be created using the FS<model>OpenHandle[] function. The function takes as arguments

  • the spectrum generator settings via the fsSettings variable
  • the Standard Model input parameters via the fsSMParameters variable
  • the model input parameters via the fsModelParameters variable

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsSettings -> { precisionGoal -> 1.*^-4 },
  fsSMParameters -> { Mt -> 173.3 },
  fsModelParameters -> {
      m0 -> 125, m12 -> 500, TanBeta -> 10, SignMu -> 1, Azero -> 0 }
];
FSCMSSMGetSettings[handle]
FSCMSSMGetSMInputParameters[handle]
FSCMSSMGetInputParameters[handle]

The FS<model>OpenHandle[] fixes all settings and input parameters at once. Unspecified parameters are set to their default values. The default values are stored in the variables fsDefaultSettings, fsDefaultSMParameters and fs<model>DefaultInputParameters:

Get["models/CMSSM/CMSSM_librarylink.m"];
Print[fsDefaultSettings];
Print[fsDefaultSMParameters];
Print[fsCMSSMDefaultInputParameters];

The settings associated to a handle can be listed using the FS<model>GetSettings[] function. Please refer to the FlexibleSUSY run-time configuration for more information on the spectrum generator settings.

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsSettings -> { precisionGoal -> 1.*^-5, betaFunctionLoopOrder -> 3 }
];
FSCMSSMGetSettings[handle]

Output:

{ precisionGoal -> 0.00001,
  maxIterations -> 0,
  calculateStandardModelMasses -> 0,
  poleMassLoopOrder -> 2,
  ewsbLoopOrder -> 2,
  betaFunctionLoopOrder -> 3,
  thresholdCorrectionsLoopOrder -> 2,
  higgs2loopCorrectionAtAs -> 1,
  higgs2loopCorrectionAbAs -> 1,
  higgs2loopCorrectionAtAt -> 1,
  higgs2loopCorrectionAtauAtau -> 1,
  forceOutput -> 0,
  top2loopCorrectionsQCD -> 1,
  betaZeroThreshold -> 1.*10^-11,
  forcePositiveMasses -> 0,
  poleMassScale -> 0.,
  parameterOutputScale -> 0. }

The Standard Model input parameters associated to a handle can be listed using the FS<model>GetSMInputParameters[] function.

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsSMParameters -> { Mt -> 173.34 }
];
FSCMSSMGetSMInputParameters[handle]

Output:

{ alphaEmMZ -> 0.00781763, (* alpha_em(MZ) in the SM(5), MS-bar *)
  GF -> 0.000011663787,    (* Fermi constant *)
  alphaSMZ -> 0.1184,      (* alpha_s(MZ) in the SM(5), MS-bar *)
  MZ -> 91.1876,           (* Z pole mass *)
  mbmb -> 4.18,            (* MS-bar bottom mass at Q = mb *)
  Mt -> 173.34,            (* top pole mass *)
  Mtau -> 1.777,           (* tau pole mass *)
  Mv3 -> 0.,               (* 3rd heaviest neutrino mass *)
  MW -> 80.385,            (* W pole mass *)
  Me -> 0.000510999,       (* electron pole mass *)
  Mv1 -> 0.,               (* 1st neutrino mass *)
  Mm -> 0.105658,          (* muon pole masss *)
  Mv2 -> 0.,               (* 2nd neutrino mass *)
  md2GeV -> 0.00475,       (* MS-bar down quark mass at Q = 2 GeV *)
  mu2GeV -> 0.0024,        (* MS-bar up quark mass at Q = 2 GeV *)
  ms2GeV -> 0.104,         (* MS-bar strange quark mass at Q = 2 GeV *)
  mcmc -> 1.27,            (* MS-bar charm quark mass at Q = mc *)
  alphaEm0 -> 0.00729735,  (* alpha_em in the Thompson limit *)
  Mh -> 125.09 }           (* Higgs pole mass *)

The model input parameters associated to a handle can be listed using the FS<model>GetInputParameters[] function.

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsModelParameters -> { m0 -> 125, m12 -> 500, TanBeta -> 10, SignMu -> 1 }
];
FSCMSSMGetInputParameters[handle]

Output:

{ m0 -> 125.,
  m12 -> 500.,
  TanBeta -> 10.,
  SignMu -> 1,
  Azero -> 0. }

Using the FS<model>Set[] function, the input parameters and settings associated to a certain handle can be modified. The FS<model>Set[] function takes first as argument the handle, and as second argument the replacement list of new parameters / settings.

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsSettings -> { precisionGoal -> 1.*^-4 },
  fsSMParameters -> { Mt -> 173.3 },
  fsModelParameters -> {
      m0 -> 125, m12 -> 500, TanBeta -> 10, SignMu -> 1, Azero -> 0 }
];

FSCMSSMGetInputParameters[handle]

FSCMSSMSet[handle, TanBeta -> 20];

FSCMSSMGetInputParameters[handle]

Output:

{m0 -> 125., m12 -> 500., TanBeta -> 10., SignMu -> 1, Azero -> 0.}

{m0 -> 125., m12 -> 500., TanBeta -> 20., SignMu -> 1, Azero -> 0.}

For each <model>, the FS<model>CalculateSpectrum[handle] function solves the boundary value problem and calculates the pole mass spectrum. The function takes a model handle as arguments, referring to the settings and input parameters

The function returns all running model parameters at the parameter output scale (either the SUSY scale or the scale set via fsSettings -> { parameterOutputScale -> 1000. }) and the running masses at the same scale. The running masses are denoted by M[p] where p is the particle name. The parameter output scale appears in the returned list with the symbol SCALE. The calculated pole masses are denoted by Pole[M[p]], respectively. The mixing matrices which correspond to the pole masses are denoted by Pole[Z], where Z is the name of the mixing matrix.

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsModelParameters -> { m0 -> 125, m12 -> 500, TanBeta -> 10, SignMu -> 1 }
];
FSCMSSMCalculateSpectrum[handle]

Output:

{CMSSM ->
   {M[VG] -> 0., M[Glu] -> 1117.18, M[Fv] -> {0., 0., 0.},
    M[Sd] -> {942.251, 977.989, 980.297, 980.3, 1023.94, 1023.94},
    M[Sv] -> {347.371, 348.42, 348.424},
    M[Su] -> {782.7, 983.889, 983.894, 987.561, 1021., 1021.},
    M[Se] -> {219.073, 226.223, 226.248, 356.971, 356.976, 358.335},
    M[hh] -> {88.1593, 732.573}, M[Ah] -> {90.0927, 732.337},
    M[Hpm] -> {78.4808, 736.531},
    M[Chi] -> {207.439, 376.528, 633.944, 647.755},
    M[Cha] -> {376.365, 647.464},
    M[Fe] -> {0.000520523, 0.107628, 1.81042},
    M[Fd] -> {0.00243143, 0.0532355, 2.32379},
    M[Fu] -> {0.00122119, 0.549091, 147.438}, M[VWm] -> 78.4808,
    M[VP] -> 0., M[VZ] -> 90.0927,
    ZD -> {{0., 0., -0.965619, 0., 0., -0.259961}, {0., 0., 0.259961, 0.,
        0., -0.965619}, {0., -0.00456672, 0., 0., -0.99999,
       0.}, {0.000208583, 0., 0., 1., 0., 0.}, {0., -0.99999, 0., 0.,
       0.00456672, 0.}, {1., 0., 0., -0.000208583, 0., 0.}},
    ZV -> {{0., 0., 1.}, {0., 1., 0.}, {1., 0., 0.}},
    ZU -> {{0., 0., 0.430138, 0., 0., 0.902763}, {0., 0.00896415, 0., 0.,
        0.99996, 0.}, {0.000019939, 0., 0., 1., 0., 0.}, {0., 0.,
       0.902763, 0., 0., -0.430138}, {1., 0., 0., -0.000019939, 0.,
       0.}, {0., 0.99996, 0., 0., -0.00896415, 0.}},
    ZE -> {{0., 0., 0.145606, 0., 0., 0.989343}, {0., -0.00903329, 0.,
       0., -0.999959, 0.}, {0.0000436949, 0., 0., 1., 0., 0.}, {1., 0.,
       0., -0.0000436949, 0., 0.}, {0., -0.999959, 0., 0., 0.00903329,
       0.}, {0., 0., 0.989343, 0., 0., -0.145606}},
    ZH -> {{0.105881, 0.994379}, {0.994379, -0.105881}},
    ZA -> {{-0.102825, 0.994699}, {0.994699, 0.102825}},
    ZP -> {{-0.102825, 0.994699}, {0.994699, 0.102825}},
    ZN -> {{-0.995744, 0.018728, -0.0832596, 0.0348113}, {0.0389752,
       0.971833, -0.194009, 0.127995}, {0. - 0.0331609 I,
       0. + 0.0485202 I, 0. + 0.703592 I,
       0. + 0.70817 I}, {0.0766551, -0.229862, -0.678518, 0.69347}},
    UM -> {{0.960661, -0.277725}, {0.277725, 0.960661}},
    UP -> {{0.983012, -0.183543}, {0.183543, 0.983012}},
    ZEL -> {{1., 0., 0.}, {0., 1., 0.}, {0., 0., 1.}},
    ZER -> {{1., 0., 0.}, {0., 1., 0.}, {0., 0., 1.}},
    ZDL -> {{1., 0., 0.}, {0., 1., 0.}, {0., 0., 1.}},
    ZDR -> {{1., 0., 0.}, {0., 1., 0.}, {0., 0., 1.}},
    ZUL -> {{1., 0., 0.}, {0., 1., 0.}, {0., 0., 1.}},
    ZUR -> {{1., 0., 0.}, {0., 1., 0.}, {0., 0., 1.}},
    ZZ -> {{-0.871112, 0.491084}, {-0.491084, -0.871112}},
    Pole[M[VG]] -> 0., Pole[M[Glu]] -> 1151.38,
    Pole[M[Fv]] -> {0., 0., 0.},
    Pole[M[Sd]] -> {970.999, 1012.32, 1015.42, 1015.42, 1059.73,
      1059.73}, Pole[M[Sv]] -> {351.491, 352.69, 352.694},
    Pole[M[Su]] -> {809.283, 1015.61, 1018.71, 1019.46, 1056.91,
      1056.91},
    Pole[M[Se]] -> {222.482, 229.821, 229.847, 361.599, 361.604,
      362.781}, Pole[M[hh]] -> {114.781, 719.259},
    Pole[M[Ah]] -> {88.5742, 718.986},
    Pole[M[Hpm]] -> {77.7605, 723.723},
    Pole[M[Chi]] -> {204.267, 385.936, 636.143, 649.77},
    Pole[M[Cha]] -> {385.949, 650.096}, Pole[M[Fe]] -> {0., 0., 0.},
    Pole[M[Fd]] -> {0., 0., 0.}, Pole[M[Fu]] -> {0., 0., 0.},
    Pole[M[VWm]] -> 80.3924, Pole[M[VP]] -> 0., Pole[M[VZ]] -> 0.,
    Pole[ZD] -> {{0., 0., -0.977566, 0., 0., -0.210631}, {0., 0.,
       0.210631, 0., 0., -0.977566}, {0., -0.0045424, 0., 0., -0.99999,
       0.}, {0.000207472, 0., 0., 1., 0., 0.}, {0., -0.99999, 0., 0.,
       0.0045424, 0.}, {1., 0., 0., -0.000207472, 0., 0.}},
    Pole[ZV] -> {{0., 0., 1.}, {0., 1., 0.}, {1., 0., 0.}},
    Pole[ZU] -> {{0., 0., 0.427999, 0., 0., 0.903779}, {0., 0., 0.903779,
        0., 0., -0.427999}, {0., 0.00911132, 0., 0., 0.999958,
       0.}, {0.0000202664, 0., 0., 1., 0., 0.}, {1., 0.,
       0., -0.0000202664, 0., 0.}, {0., 0.999958, 0., 0., -0.00911132,
       0.}}, Pole[
      ZE] -> {{0., 0., 0.144271, 0., 3.02431*10^-15,
       0.989538}, {0., -0.00895024, 2.08714*10^-14, 0., -0.99996,
       0.}, {0.0000432932, 0., 0., 1., 0., 0.}, {1., 0.,
       0., -0.0000432932, 0., 0.}, {0., -0.99996, -1.86811*10^-16, 0.,
       0.00895024, 0.}, {0., 0., -0.989538, 0., -2.06711*10^-14,
       0.144271}},
    Pole[ZH] -> {{0.106581, 0.994304}, {0.994304, -0.106581}},
    Pole[ZA] -> {{-0.0989827, 0.995089}, {0.995089, 0.0989827}},
    Pole[ZP] -> {{-0.0995943, 0.995028}, {0.995028, 0.0995943}},
    Pole[ZN] -> {{-0.995819, 0.0174686, -0.082821,
       0.0343646}, {0.0380335, 0.970567, -0.197841,
       0.131955}, {0. - 0.0332126 I, 0. + 0.0483916 I, 0. + 0.703447 I,
       0. + 0.70832 I}, {0.0761299, -0.235272, -0.677615, 0.692596}},
    Pole[UM] -> {{0.95912, -0.283001}, {0.283001, 0.95912}},
    Pole[UP] -> {{0.981917, -0.189314}, {0.189314, 0.981917}},
    Pole[ZEL] -> {{0., 0., 0.}, {0., 0., 0.}, {0., 0., 0.}},
    Pole[ZER] -> {{0., 0., 0.}, {0., 0., 0.}, {0., 0., 0.}},
    Pole[ZDL] -> {{0., 0., 0.}, {0., 0., 0.}, {0., 0., 0.}},
    Pole[ZDR] -> {{0., 0., 0.}, {0., 0., 0.}, {0., 0., 0.}},
    Pole[ZUL] -> {{0., 0., 0.}, {0., 0., 0.}, {0., 0., 0.}},
    Pole[ZUR] -> {{0., 0., 0.}, {0., 0., 0.}, {0., 0., 0.}},
    Pole[ZZ] -> {{0., 0.}, {0., 0.}},
    Yd -> {{0.000136987, 0., 0.}, {0., 0.0029993, 0.}, {0., 0.,
       0.130923}},
    Ye -> {{0.0000293264, 0., 0.}, {0., 0.00606377, 0.}, {0., 0.,
       0.102}},
    Yu -> {{7.1123*10^-6, 0., 0.}, {0., 0.00319794, 0.}, {0., 0.,
       0.858685}}, \[Mu] -> 630.611, g1 -> 0.467953, g2 -> 0.642978,
    g3 -> 1.06483, vd -> 25.1013, vu -> 242.823,
    T[Yd] -> {{-0.192259, 0., 0.}, {0., -4.20945, 0.}, {0.,
       0., -171.869}},
    T[Ye] -> {{-0.00878455, 0., 0.}, {0., -1.81633, 0.}, {0.,
       0., -30.3818}},
    T[Yu] -> {{-0.00817412, 0., 0.}, {0., -3.67535, 0.}, {0.,
       0., -764.191}}, B[\[Mu]] -> 54854.6,
    mq2 -> {{1.04513*10^6, 0., 0.}, {0., 1.04512*10^6, 0.}, {0., 0.,
       889135.}},
    ml2 -> {{125372., 0., 0.}, {0., 125369., 0.}, {0., 0., 124639.}},
    mHd2 -> 109915., mHu2 -> -385101.,
    md2 -> {{960350., 0., 0.}, {0., 960345., 0.}, {0., 0., 951180.}},
    mu2 -> {{969326., 0., 0.}, {0., 969321., 0.}, {0., 0., 659257.}},
    me2 -> {{49272.2, 0., 0.}, {0., 49266.9, 0.}, {0., 0., 47778.5}},
    MassB -> 209.358, MassWB -> 388.421, MassG -> 1117.18,
    SCALE -> 879.186}
}

For each <model>, the FS<model>CalculateObservables[handle] function calculates further observables, such as effective Higgs couplings to two photons or gluons. See the section on Observables for a list of all available observables.

Note: The FS<model>CalculateObservables[handle] function assumes, that the pole mass spectrum has been calculated before, using the FS<model>CalculateSpectrum[handle] function.

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsModelParameters -> { m0 -> 125, m12 -> 500, TanBeta -> 10, SignMu -> 1 }
];
FSCMSSMCalculateSpectrum[handle]
FSCMSSMCalculateObservables[handle]

Output:

{CMSSM ->
   { FlexibleSUSYObservable``CpHiggsPhotonPhoton ->
       {0.0000296409 - 2.1245*10^-7 I, 7.82123*10^-7 + 9.1076*10^-7 I},
     FlexibleSUSYObservable``CpHiggsGluonGluon ->
       {-0.0000670724 - 2.65658*10^-6 I, 2.72135*10^-6 + 4.91993*10^-6 I},
     FlexibleSUSYObservable``CpPseudoScalarPhotonPhoton ->
       1.05105*10^-6 - 8.33068*10^-7 I,
     FlexibleSUSYObservable``CpPseudoScalarGluonGluon ->
       6.71448*10^-6 + 8.41625*10^-7 I }
}

After the spectrum has been calculated, one should check for problems or warnings. They can be obtained for a given handle using the FS<model>GetProblems[handle] and FS<model>GetWarnings[handle] functions, respectively. These functions return the empty list if no problems / warnings occurred.

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsModelParameters -> { m0 -> 1000, m12 -> 500, Azero -> -10000, TanBeta -> 2, SignMu -> 1 }
];
FSCMSSMCalculateSpectrum[handle];
FSCMSSMGetProblems[handle]

Output:

{CMSSM ->
  { Tachyons -> {M[Sd], M[Su]},
    NoPoleMassConvergence -> {Pole[M[hh]]} }
}

This list of problems states, that the running up-type and down-type squarks are tachyonic for this parameter point. Thus, the spectrum calculated by FlexibleSUSY for this point cannot be trusted. Furthermore, the iteration to determine the Higgs pole mass did not converge. Thus, the calculated Higgs pole mass cannot be trusted either for this parameter point.

The running parameters, the mass spectrum and/or the observables can be converted to SLHA format using the FS<model>ToSLHA[handle] function. The function returns a string formatted according to [SLHA1, SLHA2].

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];
handle = FSCMSSMOpenHandle[
  fsModelParameters -> { m0 -> 1000, m12 -> 500, Azero -> 0, TanBeta -> 10, SignMu -> 1 }
];
FSCMSSMCalculateSpectrum[handle];
FSCMSSMCalculateObservables[handle];
Export["spectrum.slha", FSCMSSMToSLHA[handle], "String"];

Output: spectrum.slha

Block SPINFO
     1   FlexibleSUSY
     2   1.7.1
     5   CMSSM
     9   4.9.1
Block FlexibleSUSY
     0     1.00000000E-04   # precision goal
     1     0.00000000E+00   # max. iterations (0 = automatic)
     2     0.00000000E+00   # algorithm (0 = two_scale)
     3     0.00000000E+00   # calculate SM pole masses
     4     2.00000000E+00   # pole mass loop order
     5     2.00000000E+00   # EWSB loop order
     6     3.00000000E+00   # beta-functions loop order
     7     2.00000000E+00   # threshold corrections loop order
     8     1.00000000E+00   # Higgs 2-loop corrections O(alpha_t alpha_s)
     9     1.00000000E+00   # Higgs 2-loop corrections O(alpha_b alpha_s)
    10     1.00000000E+00   # Higgs 2-loop corrections O((alpha_t + alpha_b)^2)
    11     1.00000000E+00   # Higgs 2-loop corrections O(alpha_tau^2)
    12     0.00000000E+00   # force output
    13     1.00000000E+00   # Top pole mass QCD corrections (0 = 1L, 1 = 2L, 2 = 3L)
    14     1.00000000E-11   # beta-function zero threshold
    15     0.00000000E+00   # calculate observables (a_muon, ...)
    16     0.00000000E+00   # force positive majorana masses
    17     0.00000000E+00   # pole mass renormalization scale (0 = SUSY scale)
    18     0.00000000E+00   # pole mass renormalization scale in the EFT (0 = min(SUSY scale, Mt))
    19     0.00000000E+00   # EFT matching scale (0 = SUSY scale)
    20     2.00000000E+00   # EFT loop order for upwards matching
    21     1.00000000E+00   # EFT loop order for downwards matching
    22     0.00000000E+00   # EFT index of SM-like Higgs in the BSM model
    23     1.00000000E+00   # calculate BSM pole masses
Block SMINPUTS
     1     1.27916000E+02   # alpha^(-1) SM MSbar(MZ)
     2     1.16637000E-05   # G_Fermi
     3     1.18400000E-01   # alpha_s(MZ) SM MSbar
     4     9.11876000E+01   # MZ(pole)
     5     4.18000000E+00   # mb(mb) SM MSbar
     6     1.73340000E+02   # mtop(pole)
     7     1.77700000E+00   # mtau(pole)
     8     0.00000000E+00   # mnu3(pole)
     9     8.03850000E+01   # MW(pole)
    11     5.10998902E-04   # melectron(pole)
    12     0.00000000E+00   # mnu1(pole)
    13     1.05658372E-01   # mmuon(pole)
    14     0.00000000E+00   # mnu2(pole)
    21     4.75000000E-03   # md
    22     2.40000000E-03   # mu
    23     1.04000000E-01   # ms
    24     1.27000000E+00   # mc
Block FlexibleSUSYInput
     0     7.29735257E-03   # alpha_em(0)
     1     1.25090000E+02   # mh_pole
Block MODSEL
     6                  0   # quark/lepton flavour violation
    12     0.00000000E+00   # DRbar parameter output scale (GeV)
Block MINPAR
     1     1.00000000E+03   # m0
     2     5.00000000E+02   # m12
     3     1.00000000E+01   # TanBeta
     4                  1   # SignMu
     5     0.00000000E+00   # Azero
Block gauge Q= 1.08941472E+03
     1     3.62448909E-01   # gY
     2     6.41812871E-01   # g2
     3     1.05659964E+00   # g3
Block Yu Q= 1.08941472E+03
  1  1     7.21657623E-06   # Yu(1,1)
  1  2     0.00000000E+00   # Yu(1,2)
  1  3     0.00000000E+00   # Yu(1,3)
  2  1     0.00000000E+00   # Yu(2,1)
  2  2     3.28521207E-03   # Yu(2,2)
  2  3     0.00000000E+00   # Yu(2,3)
  3  1     0.00000000E+00   # Yu(3,1)
  3  2     0.00000000E+00   # Yu(3,2)
  3  3     8.57291451E-01   # Yu(3,3)
Block Yd Q= 1.08941472E+03
  1  1     1.38593672E-04   # Yd(1,1)
  1  2     0.00000000E+00   # Yd(1,2)
  1  3     0.00000000E+00   # Yd(1,3)
  2  1     0.00000000E+00   # Yd(2,1)
  2  2     3.03447401E-03   # Yd(2,2)
  2  3     0.00000000E+00   # Yd(2,3)
  3  1     0.00000000E+00   # Yd(3,1)
  3  2     0.00000000E+00   # Yd(3,2)
  3  3     1.31773557E-01   # Yd(3,3)
Block Ye Q= 1.08941472E+03
  1  1     2.86516322E-05   # Ye(1,1)
  1  2     0.00000000E+00   # Ye(1,2)
  1  3     0.00000000E+00   # Ye(1,3)
  2  1     0.00000000E+00   # Ye(2,1)
  2  2     5.92424993E-03   # Ye(2,2)
  2  3     0.00000000E+00   # Ye(2,3)
  3  1     0.00000000E+00   # Ye(3,1)
  3  2     0.00000000E+00   # Ye(3,2)
  3  3     9.96402101E-02   # Ye(3,3)
Block Te Q= 1.08941472E+03
  1  1    -8.51105194E-03   # TYe(1,1)
  1  2     0.00000000E+00   # TYe(1,2)
  1  3     0.00000000E+00   # TYe(1,3)
  2  1     0.00000000E+00   # TYe(2,1)
  2  2    -1.75978274E+00   # TYe(2,2)
  2  3     0.00000000E+00   # TYe(2,3)
  3  1     0.00000000E+00   # TYe(3,1)
  3  2     0.00000000E+00   # TYe(3,2)
  3  3    -2.94405366E+01   # TYe(3,3)
Block Td Q= 1.08941472E+03
  1  1    -1.88977041E-01   # TYd(1,1)
  1  2     0.00000000E+00   # TYd(1,2)
  1  3     0.00000000E+00   # TYd(1,3)
  2  1     0.00000000E+00   # TYd(2,1)
  2  2    -4.13759191E+00   # TYd(2,2)
  2  3     0.00000000E+00   # TYd(2,3)
  3  1     0.00000000E+00   # TYd(3,1)
  3  2     0.00000000E+00   # TYd(3,2)
  3  3    -1.68021836E+02   # TYd(3,3)
Block Tu Q= 1.08941472E+03
  1  1    -8.04938899E-03   # TYu(1,1)
  1  2     0.00000000E+00   # TYu(1,2)
  1  3     0.00000000E+00   # TYu(1,3)
  2  1     0.00000000E+00   # TYu(2,1)
  2  2    -3.66431893E+00   # TYu(2,2)
  2  3     0.00000000E+00   # TYu(2,3)
  3  1     0.00000000E+00   # TYu(3,1)
  3  2     0.00000000E+00   # TYu(3,2)
  3  3    -7.39719203E+02   # TYu(3,3)
Block MSQ2 Q= 1.08941472E+03
  1  1     1.92793919E+06   # mq2(1,1)
  1  2     0.00000000E+00   # mq2(1,2)
  1  3     0.00000000E+00   # mq2(1,3)
  2  1     0.00000000E+00   # mq2(2,1)
  2  2     1.92792577E+06   # mq2(2,2)
  2  3     0.00000000E+00   # mq2(2,3)
  3  1     0.00000000E+00   # mq2(3,1)
  3  2     0.00000000E+00   # mq2(3,2)
  3  3     1.46411510E+06   # mq2(3,3)
Block MSE2 Q= 1.08941472E+03
  1  1     1.02965849E+06   # me2(1,1)
  1  2     0.00000000E+00   # me2(1,2)
  1  3     0.00000000E+00   # me2(1,3)
  2  1     0.00000000E+00   # me2(2,1)
  2  2     1.02959546E+06   # me2(2,2)
  2  3     0.00000000E+00   # me2(2,3)
  3  1     0.00000000E+00   # me2(3,1)
  3  2     0.00000000E+00   # me2(3,2)
  3  3     1.01183185E+06   # me2(3,3)
Block MSL2 Q= 1.08941472E+03
  1  1     1.09811858E+06   # ml2(1,1)
  1  2     0.00000000E+00   # ml2(1,2)
  1  3     0.00000000E+00   # ml2(1,3)
  2  1     0.00000000E+00   # ml2(2,1)
  2  2     1.09808726E+06   # ml2(2,2)
  2  3     0.00000000E+00   # ml2(2,3)
  3  1     0.00000000E+00   # ml2(3,1)
  3  2     0.00000000E+00   # ml2(3,2)
  3  3     1.08926184E+06   # ml2(3,3)
Block MSU2 Q= 1.08941472E+03
  1  1     1.85944942E+06   # mu2(1,1)
  1  2     0.00000000E+00   # mu2(1,2)
  1  3     0.00000000E+00   # mu2(1,3)
  2  1     0.00000000E+00   # mu2(2,1)
  2  2     1.85943524E+06   # mu2(2,2)
  2  3     0.00000000E+00   # mu2(2,3)
  3  1     0.00000000E+00   # mu2(3,1)
  3  2     0.00000000E+00   # mu2(3,2)
  3  3     9.41968516E+05   # mu2(3,3)
Block MSD2 Q= 1.08941472E+03
  1  1     1.85122889E+06   # md2(1,1)
  1  2     0.00000000E+00   # md2(1,2)
  1  3     0.00000000E+00   # md2(1,3)
  2  1     0.00000000E+00   # md2(2,1)
  2  2     1.85121598E+06   # md2(2,2)
  2  3     0.00000000E+00   # md2(2,3)
  3  1     0.00000000E+00   # md2(3,1)
  3  2     0.00000000E+00   # md2(3,2)
  3  3     1.82729635E+06   # md2(3,3)
Block Phases Q= 1.08941472E+03
     1     1.00000000E+00   # Re(PhaseGlu)
Block IMPhases Q= 1.08941472E+03
     1     0.00000000E+00   # Im(PhaseGlu)
Block MASS
   1000021     1.19858229E+03   # Glu
        24     8.03923382E+01   # VWm
   1000024     3.92101428E+02   # Cha(1)
   1000037     6.38758479E+02   # Cha(2)
        25     1.15429842E+02   # hh(1)
        35     1.20308783E+03   # hh(2)
        37     1.20580267E+03   # Hpm(2)
        36     1.20306748E+03   # Ah(2)
   1000012     1.04330120E+03   # Sv(1)
   1000014     1.04765714E+03   # Sv(2)
   1000016     1.04767257E+03   # Sv(3)
   1000022     2.07437446E+02   # Chi(1)
   1000023     3.92113022E+02   # Chi(2)
   1000025    -6.23279987E+02   # Chi(3)
   1000035     6.38634777E+02   # Chi(4)
   1000001     1.23941353E+03   # Sd(1)
   1000003     1.38335815E+03   # Sd(2)
   1000005     1.39207702E+03   # Sd(3)
   2000001     1.39208249E+03   # Sd(4)
   2000003     1.42196160E+03   # Sd(5)
   2000005     1.42196587E+03   # Sd(6)
   1000011     1.00717444E+03   # Se(1)
   1000013     1.01697963E+03   # Se(2)
   1000015     1.01701468E+03   # Se(3)
   2000011     1.04734697E+03   # Se(4)
   2000013     1.05094581E+03   # Se(5)
   2000015     1.05095815E+03   # Se(6)
   1000002     9.94140079E+02   # Su(1)
   1000004     1.26241854E+03   # Su(2)
   1000006     1.39428015E+03   # Su(3)
   2000002     1.39428785E+03   # Su(4)
   2000004     1.41992874E+03   # Su(5)
   2000006     1.41993106E+03   # Su(6)
Block UMIX
  1  1     9.53292717E-01   # Re(UM(1,1))
  1  2    -3.02048004E-01   # Re(UM(1,2))
  2  1     3.02048004E-01   # Re(UM(2,1))
  2  2     9.53292717E-01   # Re(UM(2,2))
Block VMIX
  1  1     9.78125635E-01   # Re(UP(1,1))
  1  2    -2.08015004E-01   # Re(UP(1,2))
  2  1     2.08015004E-01   # Re(UP(2,1))
  2  2     9.78125635E-01   # Re(UP(2,2))
Block PSEUDOSCALARMIX
  1  1    -1.00061419E-01   # ZA(1,1)
  1  2     9.94981262E-01   # ZA(1,2)
  2  1     9.94981262E-01   # ZA(2,1)
  2  2     1.00061419E-01   # ZA(2,2)
Block DSQMIX
  1  1    -0.00000000E+00   # ZD(1,1)
  1  2    -0.00000000E+00   # ZD(1,2)
  1  3    -9.99033786E-01   # ZD(1,3)
  1  4    -0.00000000E+00   # ZD(1,4)
  1  5    -0.00000000E+00   # ZD(1,5)
  1  6    -4.39487765E-02   # ZD(1,6)
  2  1     0.00000000E+00   # ZD(2,1)
  2  2     0.00000000E+00   # ZD(2,2)
  2  3     4.39487765E-02   # ZD(2,3)
  2  4     0.00000000E+00   # ZD(2,4)
  2  5     0.00000000E+00   # ZD(2,5)
  2  6    -9.99033786E-01   # ZD(2,6)
  3  1     0.00000000E+00   # ZD(3,1)
  3  2     5.00403353E-03   # ZD(3,2)
  3  3     0.00000000E+00   # ZD(3,3)
  3  4     0.00000000E+00   # ZD(3,4)
  3  5     9.99987480E-01   # ZD(3,5)
  3  6     0.00000000E+00   # ZD(3,6)
  4  1     2.28556802E-04   # ZD(4,1)
  4  2     0.00000000E+00   # ZD(4,2)
  4  3     0.00000000E+00   # ZD(4,3)
  4  4     9.99999974E-01   # ZD(4,4)
  4  5     0.00000000E+00   # ZD(4,5)
  4  6     0.00000000E+00   # ZD(4,6)
  5  1     0.00000000E+00   # ZD(5,1)
  5  2     9.99987480E-01   # ZD(5,2)
  5  3     0.00000000E+00   # ZD(5,3)
  5  4     0.00000000E+00   # ZD(5,4)
  5  5    -5.00403353E-03   # ZD(5,5)
  5  6     0.00000000E+00   # ZD(5,6)
  6  1     9.99999974E-01   # ZD(6,1)
  6  2     0.00000000E+00   # ZD(6,2)
  6  3     0.00000000E+00   # ZD(6,3)
  6  4    -2.28556802E-04   # ZD(6,4)
  6  5     0.00000000E+00   # ZD(6,5)
  6  6     0.00000000E+00   # ZD(6,6)
Block SELMIX
  1  1     0.00000000E+00   # ZE(1,1)
  1  2     0.00000000E+00   # ZE(1,2)
  1  3     1.38267119E-01   # ZE(1,3)
  1  4     0.00000000E+00   # ZE(1,4)
  1  5     0.00000000E+00   # ZE(1,5)
  1  6     9.90394974E-01   # ZE(1,6)
  2  1     0.00000000E+00   # ZE(2,1)
  2  2    -9.59493336E-03   # ZE(2,2)
  2  3     0.00000000E+00   # ZE(2,3)
  2  4     0.00000000E+00   # ZE(2,4)
  2  5    -9.99953968E-01   # ZE(2,5)
  2  6     0.00000000E+00   # ZE(2,6)
  3  1     4.64326774E-05   # ZE(3,1)
  3  2     0.00000000E+00   # ZE(3,2)
  3  3     0.00000000E+00   # ZE(3,3)
  3  4     9.99999999E-01   # ZE(3,4)
  3  5     0.00000000E+00   # ZE(3,5)
  3  6     0.00000000E+00   # ZE(3,6)
  4  1     0.00000000E+00   # ZE(4,1)
  4  2     0.00000000E+00   # ZE(4,2)
  4  3     9.90394974E-01   # ZE(4,3)
  4  4     0.00000000E+00   # ZE(4,4)
  4  5     0.00000000E+00   # ZE(4,5)
  4  6    -1.38267119E-01   # ZE(4,6)
  5  1     0.00000000E+00   # ZE(5,1)
  5  2    -9.99953968E-01   # ZE(5,2)
  5  3     0.00000000E+00   # ZE(5,3)
  5  4     0.00000000E+00   # ZE(5,4)
  5  5     9.59493336E-03   # ZE(5,5)
  5  6     0.00000000E+00   # ZE(5,6)
  6  1     9.99999999E-01   # ZE(6,1)
  6  2     0.00000000E+00   # ZE(6,2)
  6  3     0.00000000E+00   # ZE(6,3)
  6  4    -4.64326774E-05   # ZE(6,4)
  6  5     0.00000000E+00   # ZE(6,5)
  6  6     0.00000000E+00   # ZE(6,6)
Block SCALARMIX
  1  1     1.04543664E-01   # ZH(1,1)
  1  2     9.94520298E-01   # ZH(1,2)
  2  1     9.94520298E-01   # ZH(2,1)
  2  2    -1.04543664E-01   # ZH(2,2)
Block NMIX
  1  1     9.95491987E-01   # Re(ZN(1,1))
  1  2    -1.79498849E-02   # Re(ZN(1,2))
  1  3     8.57113475E-02   # Re(ZN(1,3))
  1  4    -3.64289843E-02   # Re(ZN(1,4))
  2  1     4.08078351E-02   # Re(ZN(2,1))
  2  2     9.66067231E-01   # Re(ZN(2,2))
  2  3    -2.10338253E-01   # Re(ZN(2,3))
  2  4     1.44245090E-01   # Re(ZN(2,4))
  3  1    -3.37628172E-02   # Re(ZN(3,1))
  3  2     4.88174736E-02   # Re(ZN(3,2))
  3  3     7.03404869E-01   # Re(ZN(3,3))
  3  4     7.08306796E-01   # Re(ZN(3,4))
  4  1     7.86797123E-02   # Re(ZN(4,1))
  4  2    -2.52999529E-01   # Re(ZN(4,2))
  4  3    -6.73522809E-01   # Re(ZN(4,3))
  4  4     6.90049105E-01   # Re(ZN(4,4))
Block CHARGEMIX
  1  1    -1.00190943E-01   # ZP(1,1)
  1  2     9.94968228E-01   # ZP(1,2)
  2  1     9.94968228E-01   # ZP(2,1)
  2  2     1.00190943E-01   # ZP(2,2)
Block USQMIX
  1  1     0.00000000E+00   # ZU(1,1)
  1  2     0.00000000E+00   # ZU(1,2)
  1  3     2.45143979E-01   # ZU(1,3)
  1  4     0.00000000E+00   # ZU(1,4)
  1  5     7.81067721E-14   # ZU(1,5)
  1  6     9.69486683E-01   # ZU(1,6)
  2  1     0.00000000E+00   # ZU(2,1)
  2  2     0.00000000E+00   # ZU(2,2)
  2  3    -9.69486683E-01   # ZU(2,3)
  2  4     0.00000000E+00   # ZU(2,4)
  2  5    -3.08861911E-13   # ZU(2,5)
  2  6     2.45143979E-01   # ZU(2,6)
  3  1     0.00000000E+00   # ZU(3,1)
  3  2    -1.04715235E-02   # ZU(3,2)
  3  3     3.18616538E-13   # ZU(3,3)
  3  4     0.00000000E+00   # ZU(3,4)
  3  5    -9.99945172E-01   # ZU(3,5)
  3  6     0.00000000E+00   # ZU(3,6)
  4  1     2.30067962E-05   # ZU(4,1)
  4  2     0.00000000E+00   # ZU(4,2)
  4  3     0.00000000E+00   # ZU(4,3)
  4  4     1.00000000E+00   # ZU(4,4)
  4  5     0.00000000E+00   # ZU(4,5)
  4  6     0.00000000E+00   # ZU(4,6)
  5  1     0.00000000E+00   # ZU(5,1)
  5  2    -9.99945172E-01   # ZU(5,2)
  5  3    -3.33658350E-15   # ZU(5,3)
  5  4     0.00000000E+00   # ZU(5,4)
  5  5     1.04715235E-02   # ZU(5,5)
  5  6     0.00000000E+00   # ZU(5,6)
  6  1     1.00000000E+00   # ZU(6,1)
  6  2     0.00000000E+00   # ZU(6,2)
  6  3     0.00000000E+00   # ZU(6,3)
  6  4    -2.30067962E-05   # ZU(6,4)
  6  5     0.00000000E+00   # ZU(6,5)
  6  6     0.00000000E+00   # ZU(6,6)
Block SNUMIX
  1  1     0.00000000E+00   # ZV(1,1)
  1  2     0.00000000E+00   # ZV(1,2)
  1  3     1.00000000E+00   # ZV(1,3)
  2  1     0.00000000E+00   # ZV(2,1)
  2  2     1.00000000E+00   # ZV(2,2)
  2  3     0.00000000E+00   # ZV(2,3)
  3  1     1.00000000E+00   # ZV(3,1)
  3  2     0.00000000E+00   # ZV(3,2)
  3  3     0.00000000E+00   # ZV(3,3)
Block FlexibleSUSYOutput
     0     2.04206021E+16   # HighScale
     1     1.08941472E+03   # SUSYScale
     2     9.11876000E+01   # LowScale
Block FlexibleSUSYLowEnergy Q= 1.08941472E+03
    21     2.25853630E-10   # Delta(g-2)_muon/2 FlexibleSUSY 1L
Block EFFHIGGSCOUPLINGS
       25       22       22     2.99452411E-05   # Abs(effective H-Photon-Photon coupling)
       35       22       22     1.10853075E-06   # Abs(effective H-Photon-Photon coupling)
       25       21       21     6.71211022E-05   # Abs(effective H-Gluon-Gluon coupling)
       35       21       21     2.79047785E-06   # Abs(effective H-Gluon-Gluon coupling)
       36       22       22     1.73035166E-06   # Abs(effective A-Photon-Photon coupling)
       36       21       21     3.59315156E-06   # Abs(effective A-Gluon-Gluon coupling)
Block ALPHA
          -1.04735039E-01   # ArcSin(Pole(ZH(2,2)))
Block HMIX Q= 1.08941472E+03
     1     6.15787614E+02   # Mu
     2     9.64402142E+00   # vu/vd
     3     2.43666046E+02   # Sqrt(Sqr(vd) + Sqr(vu))
     4     1.47855510E+06   # Sqr(MAh(2))
   101     1.51682262E+05   # BMu
   102     2.51312780E+01   # vd
   103     2.42366584E+02   # vu
Block Au Q= 1.08941472E+03
  1  1    -1.11540275E+03   # TYu(1,1)/Yu(1,1)
  2  2    -1.11539799E+03   # TYu(2,2)/Yu(2,2)
  3  3    -8.62856153E+02   # TYu(3,3)/Yu(3,3)
Block Ad Q= 1.08941472E+03
  1  1    -1.36353297E+03   # TYd(1,1)/Yd(1,1)
  2  2    -1.36352854E+03   # TYd(2,2)/Yd(2,2)
  3  3    -1.27508007E+03   # TYd(3,3)/Yd(3,3)
Block Ae Q= 1.08941472E+03
  1  1    -2.97052953E+02   # TYe(1,1)/Ye(1,1)
  2  2    -2.97047351E+02   # TYe(2,2)/Ye(2,2)
  3  3    -2.95468431E+02   # TYe(3,3)/Ye(3,3)
Block MSOFT Q= 1.08941472E+03
     1     2.10560904E+02   # MassB
     2     3.89213480E+02   # MassWB
     3     1.10405452E+03   # MassG
    21     1.05177189E+06   # mHd2
    22    -3.46241804E+05   # mHu2
    31     1.04791153E+03   # SignedAbsSqrt(ml2(1,1))
    32     1.04789659E+03   # SignedAbsSqrt(ml2(2,2))
    33     1.04367708E+03   # SignedAbsSqrt(ml2(3,3))
    34     1.01472089E+03   # SignedAbsSqrt(me2(1,1))
    35     1.01468983E+03   # SignedAbsSqrt(me2(2,2))
    36     1.00589853E+03   # SignedAbsSqrt(me2(3,3))
    41     1.38850250E+03   # SignedAbsSqrt(mq2(1,1))
    42     1.38849767E+03   # SignedAbsSqrt(mq2(2,2))
    43     1.21000624E+03   # SignedAbsSqrt(mq2(3,3))
    44     1.36361630E+03   # SignedAbsSqrt(mu2(1,1))
    45     1.36361110E+03   # SignedAbsSqrt(mu2(2,2))
    46     9.70550625E+02   # SignedAbsSqrt(mu2(3,3))
    47     1.36059873E+03   # SignedAbsSqrt(md2(1,1))
    48     1.36059398E+03   # SignedAbsSqrt(md2(2,2))
    49     1.35177526E+03   # SignedAbsSqrt(md2(3,3))

When FlexibleSUSY is configured with --enable-verbose, a lot of additional debug output is written to stdout and stderr if FlexibleSUSY is used at the command line. When the Mathematica interface is used, this output is redirected to the notebook and printed if form of messages of type FS<model>::info, where <model> is the model name.

By default, no more than three messages of the same type are witten to the notebook. In order to write all messages to the notebook, set

Off[General::stop];

The function, which writes the messages is called FS<model>Message and is defined as

FS<model>Message[s_] := Message[FS<model>::info, s]

where s is the message string. If one would like to write the messages to a file, the function can be re-defined to

FS<model>Message[s_] := WriteString["info.txt", s <> "\n"];

Example:

Get["models/CMSSM/CMSSM_librarylink.m"];

handle = FSCMSSMOpenHandle[
  fsModelParameters -> { m0 -> 125, m12 -> 500, TanBeta -> 10, SignMu -> 1 }
];

(* write all messages to "info.txt" *)
FSCMSSMMessage[s_] := WriteString["info.txt", s <> "\n"];

FSCMSSMCalculateSpectrum[handle]