public static Mode[] GetModes(MatrixByArr hess, string cachepath = null)
{
Vector[] eigvec;
double[] eigval;
if (cachepath != null && HFile.Exists(cachepath))
{
HSerialize.Deserialize(cachepath, null, out eigval, out eigvec);
}
else
{
HDebug.Verify(NumericSolver.Eig(hess, out eigvec, out eigval));
HSerialize.SerializeDepreciated(cachepath, null, eigval, eigvec);
}
List <Mode> modes;
{ // sort by eigenvalues
int[] idx = eigval.HAbs().HIdxSorted();
modes = new List <Mode>(idx.Length);
for (int i = 0; i < eigval.Length; i++)
{
Mode mode = new Mode {
eigval = eigval[idx[i]],
eigvec = eigvec[idx[i]]
};
modes.Add(mode);
}
}
return(modes.ToArray());
}
public static Anisou[] FromHessian(Mode[] modesMassWeighted, double[] mass, double scale = 10000 *1000
, string cachepath = null
)
{
if (cachepath != null && HFile.Exists(cachepath))
{
List <Anisou> lstanisou;
HDebug.Verify(HSerialize.Deserialize(cachepath, null, out lstanisou));
return(lstanisou.ToArray());
}
int size = modesMassWeighted.Size();
HDebug.Assert(size == mass.Length);
Anisou[] anisous = new Anisou[size];
for (int i = 0; i < size; i++)
{
MatrixByArr invHii = new double[3, 3];
foreach (Mode mode in modesMassWeighted)
{
Vector modei = mode.GetEigvecOfAtom(i);
invHii = invHii + LinAlg.VVt(modei, modei) * mode.eigval;
}
MatrixByArr Ui = invHii * scale / mass[i];
anisous[i] = Anisou.FromMatrix(Ui);
}
if (cachepath != null)
{
HSerialize.Serialize(cachepath, null, new List <Anisou>(anisous));
}
return(anisous);
}
public static Anisou[] FromHessian(MatrixByArr hessMassWeighted, double[] mass, double scale = 10000 *1000
, string cachepath = null
)
{
/// Estimation of "anisotropic temperature factors" (ANISOU)
///
/// delta = hess^-1 * force
/// = (0 + V7*V7'/L7 + V8*V8'/L8 + V9*V9'/L9 + ...) * force (* assume that 1-6 eigvecs/eigvals are ignored, because rot,trans *)
///
/// Assume that force[i] follows gaussian distributions N(0,1). Here, if there are 1000 samples, let denote i-th force as fi, and its j-th element as fi[j]
/// Then, $V7' * fi = si7, V8' * fi = si8, ...$ follows gaussian distribution N(0,1), too.
/// Its moved position by k-th eigen component is determined then, as
/// dik = (Vk * Vk' / Lk) * Fi
/// = Vk / Lk * (Vk' * Fi)
/// = Vk / Lk * Sik.
/// Additionally, the moved position j-th atom is:
/// dik[j] = Vk[j] / Lk[j] * Sik.
/// and its correlation matrix is written as (because its mean position is 0 !!!):
/// Cik[j] = dik[j] * dik[j]'
/// = [dik[j]_x * dik[j]_x dik[j]_x * dik[j]_y dik[j]_x * dik[j]_z]
/// [dik[j]_y * dik[j]_x dik[j]_y * dik[j]_y dik[j]_y * dik[j]_z]
/// [dik[j]_z * dik[j]_x dik[j]_z * dik[j]_y dik[j]_z * dik[j]_z]
/// = (Vk[j] * Vk[j]') / (Lk[j]*Lk[j]) * (Sik*Sik).
///
/// Note that Sik*Sik follows the chi-square distribution, because Sik follows the gaussian distribution N(0,1).
/// Additionally, note that the thermal fluctuation is (not one projection toward k-th eigen component with only i-th force, but) the results of 1..i.. forced movements and 1..k.. eigen components.
/// Therefore, for j-th atom, the accumulation of the correlation over all forces (1..i..) with all eigen components (1..k..) is:
/// C[j] = sum_{i,k} {(Vk[j] * Vk[j]') / (Lk[j]*Lk[j]) * (Sik*Sik)}.
///
/// Here, Sik is normal distribution independent to i and k. Therefore, the mean of C[j] is
/// E(C[j]) = E( sum_{i,k} {(Vk[j] * Vk[j]') / (Lk[j]*Lk[j]) * (Sik*Sik)} )
/// = sum_{i,k} E( (Vk[j] * Vk[j]') / (Lk[j]*Lk[j]) * (Sik*Sik) )
/// = sum_{i,k} { (Vk[j] * Vk[j]') / (Lk[j]*Lk[j]) * E(Sik*Sik) }
/// = sum_{i,k} { (Vk[j] * Vk[j]') / (Lk[j]*Lk[j]) * 1 } (* because mean of E(x*x)=1 where x~N(0,1) *)
/// = sum_{k} { (Vk[j] * Vk[j]') / (Lk[j]*Lk[j]) }
///
/// Note that E(C[j]) is same to the j-th diagonal component of inverse hessian matrix (except, the eigenvalues are squared).
///
/// Fixation: Gromacx generate the ensemble X by
/// X[j] = sum_{k} {Vk / sqrt(Lk[j]) / sqrt(mass[j]) * x_k},
/// where x~N(0,1). However, the above is assumed as
/// X[j] = sum_{k} {Vk / Lk[j] * x_k}.
/// In order to apply the assumption by the Gromacs ensemble, The equation should be fixed as
/// E(C[j]) = sum_{k} { (Vk[j] * Vk[j]') / (sqrt(Lk[j])*sqrt(Lk[j])) }
/// = sum_{k} { (Vk[j] * Vk[j]') / Lk[j] / mass[j] }
///
int size = mass.Length;
HDebug.Assert(hessMassWeighted.RowSize == size * 3, hessMassWeighted.ColSize == size * 3);
Anisou[] anisous = new Anisou[size];
if (cachepath != null && HFile.Exists(cachepath))
{
List <Anisou> lstanisou;
HDebug.Verify(HSerialize.Deserialize <List <Anisou> >(cachepath, null, out lstanisou));
anisous = lstanisou.ToArray();
return(anisous);
}
// anisotropic temperature factors
using (new Matlab.NamedLock("ANISOU"))
{
Matlab.Clear("ANISOU");
Matlab.PutMatrix("ANISOU.H", hessMassWeighted);
Matlab.Execute("[ANISOU.V,ANISOU.D] = eig(ANISOU.H);");
Matlab.Execute("ANISOU.D = diag(ANISOU.D);"); // get diagonal
{
Matlab.Execute("[ANISOU.sortD, ANISOU.sortIdxD] = sort(abs(ANISOU.D));"); // sorted index of abs(D)
Matlab.Execute("ANISOU.D(ANISOU.sortIdxD(1:6)) = 0;"); // set the 6 smallest eigenvalues as zero
//Matlab.Execute("ANISOU.D(ANISOU.D < 0) = 0;"); // set negative eigenvalues as zero
}
//{
// Matlab.Execute("ANISOU.D(1:6) = 0;");
//}
Matlab.Execute("ANISOU.invD = 1 ./ ANISOU.D;"); // set invD
Matlab.Execute("ANISOU.invD(ANISOU.D == 0) = 0;"); // set Inf (by divided by zero) as zero
//Matlab.Execute("ANISOU.D = ANISOU.D .^ 2;"); // assume the gromacs ensemble condition
Matlab.Execute("ANISOU.invH = ANISOU.V * diag(ANISOU.invD) * ANISOU.V';");
for (int i = 0; i < size; i++)
{
string idx = string.Format("{0}:{1}", i * 3 + 1, i * 3 + 3);
MatrixByArr U = Matlab.GetMatrix("ANISOU.invH(" + idx + "," + idx + ")");
U *= (scale / mass[i]);
anisous[i] = Anisou.FromMatrix(U);
}
Matlab.Clear("ANISOU");
}
if (cachepath != null)
{
HSerialize.Serialize(cachepath, null, new List <Anisou>(anisous));
}
return(anisous);
}
public static PdbInfo[] GetPdbInfo(params string[] pdbids)
{
Dictionary <string, PdbInfo> pdbinfos = new Dictionary <string, PdbInfo>();
int VER = 0;
string cachepath = RootPath + @"cache\GetPdbInfo.data";
if (HFile.Exists(cachepath))
{
HSerialize.Deserialize(cachepath, VER, out pdbinfos);
if (pdbinfos == null)
{
pdbinfos = new Dictionary <string, PdbInfo>();
}
}
bool updated = false;
for (int i = 0; i < pdbids.Length; i++)
{
string pdbid = pdbids[i];
if (pdbinfos.ContainsKey(pdbid) == false)
{
pdbinfos.Add(pdbid, null);
}
if (pdbinfos[pdbid] != null)
{
continue;
}
updated = true;
//continue;
List <string> pdblines = GetPdbLines(pdbid);
if (pdblines == null)
{
pdblines = GetPdbLines(pdbid, forceToRedownload: true);
}
HDebug.Assert(pdblines != null);
if (pdblines == null)
{
System.Console.WriteLine(pdbid + " is not processed");
continue;
}
PdbInfo pdbinfo = GetPdbInfo(pdbid, pdblines);
pdbinfos[pdbid] = pdbinfo;
if (i % 10 == 0)
{
System.Console.WriteLine(pdbid + " is processed. There are " + (pdbids.Length - i).ToString() + " unprocessed pdbs.");
}
if (i % 200 == 0)
{
HSerialize.Serialize(cachepath, VER, pdbinfos);
updated = false;
System.Console.WriteLine("serialize cache");
}
}
//GetPdbInfo(pdbinfos);
if (updated)
{
HSerialize.Serialize(cachepath, VER, pdbinfos);
}
return(pdbinfos.HSelectByKeys(pdbids));
}
public static void SelfTest()
{
if (HDebug.Selftest() == false)
{
return;
}
string temppath = @"K:\temp\";
string tinkerpath_testgrad = "\"" + @"C:\Program Files\Tinker\bin-win64-8.2.1\testgrad.exe" + "\"";
string tinkerpath_testhess = "\"" + @"C:\Program Files\Tinker\bin-win64-8.2.1\testhess.exe" + "\"";
var xyz = Tinker.Xyz.FromLines(SelftestData.lines_1L2Y_xyz);
var prm = Tinker.Prm.FromLines(SelftestData.lines_charmm22_prm);
var univ = Universe.Build(xyz, prm);
var testhess = Tinker.Run.Testhess(tinkerpath_testhess, xyz, prm, temppath
, HessMatrixZeros: HessMatrixLayeredArray.ZerosHessMatrixLayeredArray
);
var testgrad = Tinker.Run.Testgrad(tinkerpath_testgrad, xyz, prm, temppath);
var hessinfo = Hess.HessInfo.FromTinker(xyz, prm, testhess.hess);
var hessforcinfo = HessForc.Coarse.HessForcInfo.From(hessinfo);
hessforcinfo.forc = testgrad.anlyts.GetForces(xyz.atoms);
var coarseinfo_debug = HessForc.Coarse.GetCoarseHessForc
(hessforcinfo
, coords: hessinfo.coords
, GetIdxKeepListRemv: GetIdxKeepListRemv
, ila: null
, thres_zeroblk: double.Epsilon
, options: new string[] { "Debug" }
);
var coarseinfo_simple = HessForc.Coarse.GetCoarseHessForc
(hessforcinfo
, coords: hessinfo.coords
, GetIdxKeepListRemv: GetIdxKeepListRemv
, ila: null
, thres_zeroblk: double.Epsilon
, options: new string[] { "SubSimple" }
);
double absmax_simple = (coarseinfo_debug.hess - coarseinfo_simple.hess).HAbsMax();
HDebug.Assert(Math.Abs(absmax_simple) < 0.00000001);
double absmax_simple_forc = (coarseinfo_debug.forc.ToVector() - coarseinfo_simple.forc.ToVector()).ToArray().MaxAbs();
HDebug.Assert(Math.Abs(absmax_simple_forc) < 0.00000001);
var coarseinfo_1iter = HessForc.Coarse.GetCoarseHessForc
(hessforcinfo
, coords: hessinfo.coords
, GetIdxKeepListRemv: GetIdxKeepListRemv
, ila: null
, thres_zeroblk: double.Epsilon
, options: new string[] { "OneIter" }
);
double absmax_1iter = (coarseinfo_debug.hess - coarseinfo_1iter.hess).HAbsMax();
HDebug.Assert(Math.Abs(absmax_1iter) < 0.00000001);
double absmax_1iter_forc = (coarseinfo_debug.forc.ToVector() - coarseinfo_1iter.forc.ToVector()).ToArray().MaxAbs();
HDebug.Assert(Math.Abs(absmax_1iter_forc) < 0.00000001);
var coarseinfo_iter = HessForc.Coarse.GetCoarseHessForc
(hessforcinfo
, coords: hessinfo.coords
, GetIdxKeepListRemv: GetIdxKeepListRemv
, ila: null
, thres_zeroblk: double.Epsilon
, options: null
);
double absmax_iter = (coarseinfo_debug.hess - coarseinfo_iter.hess).HAbsMax();
HDebug.Assert(Math.Abs(absmax_iter) < 0.00000001);
double absmax_iter_forc = (coarseinfo_debug.forc.ToVector() - coarseinfo_iter.forc.ToVector()).ToArray().MaxAbs();
HDebug.Assert(Math.Abs(absmax_iter_forc) < 0.00000001);
double tolerance = 1.0E-6; // 0.00001;
var coarseinfo_1iter_tolerant = HessForc.Coarse.GetCoarseHessForc
(hessforcinfo
, coords: hessinfo.coords
, GetIdxKeepListRemv: GetIdxKeepListRemv
, ila: null
, thres_zeroblk: tolerance
, options: new string[] { "OneIter" }
);
double absmax_1iter_tolerant = (coarseinfo_debug.hess - coarseinfo_1iter_tolerant.hess).HAbsMax();
HDebug.Assert(Math.Abs(absmax_1iter_tolerant) < tolerance * 10);
double absmax_1iter_tolerant_forc = (coarseinfo_debug.forc.ToVector() - coarseinfo_1iter_tolerant.forc.ToVector()).ToArray().MaxAbs();
HDebug.Assert(Math.Abs(absmax_1iter_tolerant_forc) < tolerance * 10);
var coarseinfo_iter_tolerant = HessForc.Coarse.GetCoarseHessForc
(hessforcinfo
, coords: hessinfo.coords
, GetIdxKeepListRemv: GetIdxKeepListRemv
, ila: null
, thres_zeroblk: tolerance
, options: null
);
double absmax_iter_tolerant = (coarseinfo_debug.hess - coarseinfo_iter_tolerant.hess).HAbsMax();
HDebug.Assert(Math.Abs(absmax_iter_tolerant) < tolerance * 10);
double absmax_iter_tolerant_forc = (coarseinfo_debug.forc.ToVector() - coarseinfo_iter_tolerant.forc.ToVector()).ToArray().MaxAbs();
HDebug.Assert(Math.Abs(absmax_iter_tolerant_forc) < tolerance * 10);
string tempfilepath = HFile.GetTempPath(temppath, "test_serialzation_CoarseHessForc.dat");
HSerialize.Serialize(tempfilepath, null, coarseinfo_iter_tolerant);
var coarseinfo_iter_tolerant2 = HSerialize.Deserialize <HessForcInfo>(tempfilepath, null);
double absmax_iter_tolerant_file = (coarseinfo_iter_tolerant.hess - coarseinfo_iter_tolerant.hess).HAbsMax();
HDebug.Assert(Math.Abs(absmax_iter_tolerant_file) == 0);
double absmax_iter_tolerant_file_forc = (coarseinfo_iter_tolerant.forc.ToVector() - coarseinfo_iter_tolerant.forc.ToVector()).ToArray().MaxAbs();
HDebug.Assert(Math.Abs(absmax_iter_tolerant_file_forc) == 0);
HFile.Delete(tempfilepath);
}
public void LoadCoords(string path)
{
Vector[] coords;
HSerialize.Deserialize(path, null, out coords);
SetCoords(coords);
}
