/*

PHYML :  a program that  computes maximum likelihood  phylogenies from
DNA or AA homologous sequences

Copyright (C) Stephane Guindon. Oct 2003 onward

All parts of  the source except where indicated  are distributed under
the GNU public licence.  See http://www.opensource.org for details.

*/

/*

Implementation of aLRT branch tests, and 5-branchs NNI research optimization.

Authors : Jean-Francois Dufayard & Stephane Guindon.

*/

#include "alrt.h"

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////


/*
* Check every testable branch of the tree,
* check for NNIs, optimizing 5 branches,
* param tree : the tree to check
*/

int Check_NNI_Five_Branches(t_tree *tree)
{
  int best_edge;
  phydbl best_gain;
  int best_config;
  int i;
  int better_found; /* = 1 if a phylogeny with greater likelihood than current one was found */
  int result;
  phydbl init_lnL;

  init_lnL     = UNLIKELY;
  better_found = 1;

  //While there is at least one NNI to do...
  while(better_found)
    {
      Update_Dirs(tree);

      //Interface output
      if((tree->mod->s_opt->print) && (!tree->io->quiet)) PhyML_Printf("\n\n. Checking for NNIs, optimizing five branches...\n");

      better_found  =  0;
      result        = -1;
      best_edge     = -1;
      best_gain     = 1.E+10;
      best_config   = -1;

      MIXT_Set_Alias_Subpatt(YES,tree);
      Set_Both_Sides(YES,tree);     
      init_lnL = Lk(NULL,tree);
      MIXT_Set_Alias_Subpatt(NO,tree);

      //For every branch
      For(i,2*tree->n_otu-3)
	{
	  //if this branch is not terminal
	  if((!tree->a_edges[i]->left->tax) && (!tree->a_edges[i]->rght->tax))
	    {
	      result = -1;

	      //Try every NNIs for the tested branch
	      result = NNI_Neigh_BL(tree->a_edges[i],tree);

	      //Look for possible NNI to do, and check if it is the best one
	      switch(result)
		{
		case 1 : /* lk1 > lk0 > lk2 */
		  {
		    if((tree->a_edges[i]->nni->lk0 - tree->a_edges[i]->nni->lk1) < best_gain)
		      {
			better_found = 1;
			best_edge    = i;
			best_gain    = tree->a_edges[i]->nni->lk0-tree->a_edges[i]->nni->lk1;
			best_config  = 1;
		      }
		    break;
		  }
		case 2 : /* lk2 > lk0 > lk1 */
		  {
		    if((tree->a_edges[i]->nni->lk0 - tree->a_edges[i]->nni->lk2) < best_gain)
		      {
 			better_found = 1;
			best_edge    = i;
			best_gain    = tree->a_edges[i]->nni->lk0-tree->a_edges[i]->nni->lk2;
			best_config  = 2;
		      }
		    break;
		  }
		case 3 : /* lk1 > lk2 > lk0 */
		  {
		    if((tree->a_edges[i]->nni->lk2 - tree->a_edges[i]->nni->lk1) < best_gain)
		      {
 			better_found = 1;
			best_edge    = i;
			best_gain    = tree->a_edges[i]->nni->lk2-tree->a_edges[i]->nni->lk1;
			best_config  = 1;
		      }
		    break;
		  }
		case 4 : /* lk2 > lk1 > lk0 */
		  {
		    if((tree->a_edges[i]->nni->lk1 - tree->a_edges[i]->nni->lk2) < best_gain)
		      {
			better_found = 1;
			best_edge    = i;
			best_gain    = tree->a_edges[i]->nni->lk1-tree->a_edges[i]->nni->lk2;
			best_config  = 2;
		      }
		    break;
		  }
		default : /* lk2 = lk1 = lk0 */
		  {
		    if(best_gain > .0) best_gain = .0;
		    break;
		  }
		}
	    }
	}

      if((tree->c_lnL < init_lnL - tree->mod->s_opt->min_diff_lk_local) || (tree->c_lnL > init_lnL + tree->mod->s_opt->min_diff_lk_local))
	{
	  PhyML_Printf("\n\n== tree->c_lnL = %f init_lnL = %f.",tree->c_lnL,init_lnL);
	  PhyML_Printf("\n== Err in file %s at line %d\n\n.\n",__FILE__,__LINE__);
	  Warn_And_Exit("\n");
	}

      //Don't do any NNI if the user doesn't want to optimize topology
      if(!tree->mod->s_opt->opt_topo) better_found = 0;
/*       if(FABS(best_gain) <= tree->mod->s_opt->min_diff_lk_move) better_found = 0; */

      //If there is one swap to do, make the best one.
      if(better_found)
	{
	  Make_Target_Swap(tree,tree->a_edges[best_edge],best_config);

          MIXT_Set_Alias_Subpatt(YES,tree);
	  Lk(NULL,tree);
          MIXT_Set_Alias_Subpatt(YES,tree);

	  if(tree->c_lnL < init_lnL)
	    {
	      PhyML_Printf("\n\n== tree->c_lnL = %f init_lnL = %f.",tree->c_lnL,init_lnL);
	      PhyML_Printf("\n== Err. in file %s at line %d\n\n.\n",__FILE__,__LINE__);
	      Exit("\n");
	    }

	  if((tree->mod->s_opt->print) && (!tree->io->quiet)) Print_Lk(tree,"[Topology           ]");

	  if(FABS(tree->c_lnL - init_lnL) < tree->mod->s_opt->min_diff_lk_move) return 0;	  
	  return 1;
	}
    }
  return 0;
}

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/* Compute aLRT supports */
void aLRT(t_tree *tree)
{
  int i;
  char *method;

  Unscale_Br_Len_Multiplier_Tree(tree);
  Br_Len_Not_Involving_Invar(tree);

  /* aLRT support will label each internal branch */
  tree->print_alrt_val = YES;

  /* The topology will not be modified when assessing the branch support. We make sure that it will
     not be modified afterwards by locking the topology */

  method = (char *)mCalloc(100,sizeof(char));

  switch(tree->io->ratio_test)
    {
    case ALRTCHI2:      { strcpy(method,"aLRT"); break; }
    case MINALRTCHI2SH: { strcpy(method,"aLRT"); break; }
    case ALRTSTAT:      { strcpy(method,"aLRT"); break; }
    case SH:            { strcpy(method,"SH"); break; }
    case ABAYES:        { strcpy(method,"aBayes"); break; }
    default : return;
    }

  if(tree->io->quiet == NO) PhyML_Printf("\n\n. Calculating fast branch supports (using '%s').",method);
  Free(method);

  MIXT_Set_Alias_Subpatt(YES,tree);
  Set_Both_Sides(YES,tree);     
  Lk(NULL,tree);
  MIXT_Set_Alias_Subpatt(NO,tree);

  For(i,2*tree->n_otu-3)
    if((!tree->a_edges[i]->left->tax) && (!tree->a_edges[i]->rght->tax)) 
      {
	/* Compute likelihoods for each of the three configuration */
       	NNI_Neigh_BL(tree->a_edges[i],tree);
	/* Compute the corresponding statistical support */
	Compute_Likelihood_Ratio_Test(tree->a_edges[i],tree);
      }
  
  tree->lock_topo = YES;

  Br_Len_Involving_Invar(tree);
  Rescale_Br_Len_Multiplier_Tree(tree);

}

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/*
* Launch one branch testing,
* analyse the result
* and convert supports as wished by the user.
* param tree : the tree to check
* param tested_t_edge : the tested t_edge of the tree
* param old_loglk : the initial likelihood, before any aLRT analysis
* param isBoot : 1 if used from the Bootstrap procedure, 0 if not
* return an integer, informative to analyse the results and potential NNIs to do
*/
int Compute_Likelihood_Ratio_Test(t_edge *tested_edge, t_tree *tree)
{
  int result=0;

  tested_edge->ratio_test     =  0.0;
  tested_edge->alrt_statistic = -1.0;
  
  if((tested_edge->nni->lk0 > tested_edge->nni->lk1) && (tested_edge->nni->lk0 > tested_edge->nni->lk2))
    {
      if(tested_edge->nni->lk1 < tested_edge->nni->lk2)
	{
	  //lk0 > lk2 > lk1
	  tested_edge->alrt_statistic = 2*(tested_edge->nni->lk0 - tested_edge->nni->lk2);
	}
      else
	{
	  //lk0 > lk1 >= lk2
	  tested_edge->alrt_statistic = 2*(tested_edge->nni->lk0 - tested_edge->nni->lk1);
	}

      if (tested_edge->alrt_statistic < 0.0)
	{
	  tested_edge->alrt_statistic = 0.0;
	  tested_edge->ratio_test = 0.0;
	}
      else
	{
	  //aLRT statistic is valid, compute the branch support
	  if (tree->io->ratio_test == ALRTCHI2)
	    {
	      tested_edge->ratio_test = Statistics_To_Probabilities(tested_edge->alrt_statistic);	  
	    }
	  else if(tree->io->ratio_test == MINALRTCHI2SH)
	    {
	      phydbl sh_support;
	      phydbl param_support;
	      
	      sh_support    = Statistics_To_SH(tree);
	      param_support = Statistics_To_Probabilities(tested_edge->alrt_statistic);
	      
	      if(sh_support < param_support) tested_edge->ratio_test = sh_support;
	      else                           tested_edge->ratio_test = param_support;
	    }
	  
	  else if(tree->io->ratio_test == ALRTSTAT) 
	    {
	      tested_edge->ratio_test=tested_edge->alrt_statistic;
	    } 
	  else if(tree->io->ratio_test == SH) 
	    {
	      tested_edge->ratio_test = Statistics_To_SH(tree);
	    }
	  else if(tree->io->ratio_test == ABAYES)
	    {
	      phydbl Kp0,Kp1,Kp2,logK;
	      
	      logK = 1. - MAX(MAX(tested_edge->nni->lk0,tested_edge->nni->lk1),tested_edge->nni->lk2);

	      Kp0 = EXP(tested_edge->nni->lk0+logK);
	      Kp1 = EXP(tested_edge->nni->lk1+logK);
	      Kp2 = EXP(tested_edge->nni->lk2+logK);
	      tested_edge->ratio_test = Kp0/(Kp0+Kp1+Kp2);
	      
	    }
	}
    }  
  //statistic is not valid, give the negative statistics if wished, or a zero support.
  else if((tested_edge->nni->lk1 > tested_edge->nni->lk0) && (tested_edge->nni->lk1 > tested_edge->nni->lk2))
    {
/*       tested_edge->ratio_test = 2*(tested_edge->nni->lk0-tested_edge->nni->lk1); */
      tested_edge->ratio_test = 0.0;
      if(tree->io->ratio_test > 1) tested_edge->alrt_statistic = 0.0;
    }
  else if((tested_edge->nni->lk2 > tested_edge->nni->lk0) && (tested_edge->nni->lk2 > tested_edge->nni->lk1))
    {
/*       tested_edge->ratio_test = 2*(tested_edge->nni->lk0-tested_edge->nni->lk2); */
      tested_edge->ratio_test = 0.0;
      if(tree->io->ratio_test > 1) tested_edge->alrt_statistic = 0.0;
    }
  else // lk0 ~ lk1 ~ lk2
    {
      tested_edge->ratio_test = 0.0;
      if(tree->io->ratio_test > 1) tested_edge->ratio_test = 0.0;
    }
  return result;
}

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/*
* Test the 3 NNI positions for one branch.
* param tree : the tree to check
* param tested_t_edge : the tested t_edge of the tree
* param old_loglk : the initial likelihood, before any aLRT analysis
* param isBoot : 1 if used from the Bootstrap procedure, 0 if not
*/
int NNI_Neigh_BL(t_edge *b_fcus, t_tree *tree)
{
  int site;
  t_node *v1,*v2,*v3,*v4;
  t_edge *e1,*e2,*e3,*e4;
  phydbl *len_e1,*len_e2,*len_e3,*len_e4;
  phydbl lk0, lk1, lk2;
  phydbl *bl_init;
  phydbl l_infa, l_infb;
  phydbl lk_init, lk_temp;
  int i;
  int result,counter,wei;
  void *buff;

  if(tree->n_root || tree->e_root)
    {
      PhyML_Printf("\n== Err. in file %s at line %d\n\n",__FILE__,__LINE__);
      Exit("");
    }

  result = 0;

  /*! Initialization */
  bl_init = MIXT_Get_Lengths_Of_This_Edge(b_fcus,tree);
  lk_init = tree->c_lnL;
  lk_temp = UNLIKELY;

  b_fcus->nni->score = .0;

  lk0 = lk1 = lk2    = UNLIKELY;
  v1 = v2 = v3 = v4  = NULL;

  /*! vertices */
  v1 = b_fcus->left->v[b_fcus->l_v1];
  v2 = b_fcus->left->v[b_fcus->l_v2];
  v3 = b_fcus->rght->v[b_fcus->r_v1];
  v4 = b_fcus->rght->v[b_fcus->r_v2];

  if(v1->num < v2->num)
    {
      PhyML_Printf("\n== Err. in file %s at line %d\n\n",__FILE__,__LINE__);
      Exit("");
    }
  if(v3->num < v4->num)
    {
      PhyML_Printf("\n== Err. in file %s at line %d\n\n",__FILE__,__LINE__);
      Exit("");
    }

  /*! edges */
  e1 = b_fcus->left->b[b_fcus->l_v1];
  e2 = b_fcus->left->b[b_fcus->l_v2];
  e3 = b_fcus->rght->b[b_fcus->r_v1];
  e4 = b_fcus->rght->b[b_fcus->r_v2];

  /*! record initial branch lengths */
  len_e1 = MIXT_Get_Lengths_Of_This_Edge(e1,tree);
  len_e2 = MIXT_Get_Lengths_Of_This_Edge(e2,tree);
  len_e3 = MIXT_Get_Lengths_Of_This_Edge(e3,tree);
  len_e4 = MIXT_Get_Lengths_Of_This_Edge(e4,tree);

  /*! Optimize branch lengths and update likelihoods for
    the original configuration.
  */
  do
    {
      lk0 = lk_temp;

      For(i,3)
	if(b_fcus->left->v[i] != b_fcus->rght)
	  {
	    Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);

	    l_infa  = 10.;
	    l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->left->b[i]->l->v);

	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->left->b[i],tree);
	  }

      Update_P_Lk(tree,b_fcus,b_fcus->left);

      l_infa  = 10.;
      l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->l->v);
      lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus,tree);


      For(i,3)
	if(b_fcus->rght->v[i] != b_fcus->left)
	  {
	    Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);

	    l_infa  = 10.;
	    l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->rght->b[i]->l->v);

	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->rght->b[i],tree);
	  }

      Update_P_Lk(tree,b_fcus,b_fcus->rght);

      if(lk_temp < lk0 - tree->mod->s_opt->min_diff_lk_local)
	{
	  PhyML_Printf("\n== lk_temp = %f lk0 = %f\n",lk_temp,lk0);
	  PhyML_Printf("\n== Err. in file %s at line %d\n\n",__FILE__,__LINE__);
	  Exit("\n");
	}
    }
  while(FABS(lk_temp-lk0) > tree->mod->s_opt->min_diff_lk_global);

  lk0 = tree->c_lnL;

  buff = (t_tree *)tree;
  do
    {      
      counter=0;
      For(site,tree->n_pattern)
        {
          wei=0;
          For(wei,tree->data->wght[site])
            {
              tree->log_lks_aLRT[0][counter]= tree->c_lnL_sorted[site] / tree->data->wght[site];
              counter++;
            }
        }
      tree = tree->next_mixt;
    }
  while(tree);
  tree = (t_tree *)buff;


  /*! Go back to initial branch lengths */
  MIXT_Set_Lengths_Of_This_Edge(len_e1,e1,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e2,e2,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e3,e3,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e4,e4,tree);
  MIXT_Set_Lengths_Of_This_Edge(bl_init,b_fcus,tree);
  Update_PMat_At_Given_Edge(e1,tree);
  Update_PMat_At_Given_Edge(e2,tree);
  Update_PMat_At_Given_Edge(e3,tree);
  Update_PMat_At_Given_Edge(e4,tree);
  Update_PMat_At_Given_Edge(b_fcus,tree);


  /* Sanity check */
  MIXT_Set_Alias_Subpatt(YES,tree);
  Update_P_Lk(tree,b_fcus,b_fcus->rght);
  Update_P_Lk(tree,b_fcus,b_fcus->left);
  lk_temp = Lk(b_fcus,tree);
  MIXT_Set_Alias_Subpatt(NO,tree);
  if((lk_temp > lk_init + tree->mod->s_opt->min_diff_lk_local) || (lk_temp < lk_init - tree->mod->s_opt->min_diff_lk_local))
    {
      PhyML_Printf("\n== lk_temp = %f lk_init = %f",lk_temp,lk_init);
      PhyML_Printf("\n== Err. in file %s at line %d\n",__FILE__,__LINE__);
      Exit("\n");
    }


  /*! Do first possible swap */
  Swap(v2,b_fcus->left,b_fcus->rght,v3,tree);

  MIXT_Set_Alias_Subpatt(YES,tree);
  Set_Both_Sides(YES,tree);     
  Update_P_Lk(tree,b_fcus,b_fcus->left);
  Update_P_Lk(tree,b_fcus,b_fcus->rght);
  lk1 = Lk(b_fcus,tree);
  MIXT_Set_Alias_Subpatt(NO,tree);

  For(i,3)
    if(b_fcus->left->v[i] != b_fcus->rght)
      Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);

  For(i,3)
    if(b_fcus->rght->v[i] != b_fcus->left)
      Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);


  /*! Optimize branch lengths and update likelihoods */
  lk_temp = UNLIKELY;
  do
    {
      lk1 = lk_temp;

      For(i,3)
	if(b_fcus->left->v[i] != b_fcus->rght)
	  {
	    Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);
	    l_infa  = 10.;
	    l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->left->b[i]->l->v);
	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->left->b[i],tree);
	  }

      Update_P_Lk(tree,b_fcus,b_fcus->left);
      l_infa  = 10.;
      l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->l->v);
      lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus,tree);

      For(i,3)
	if(b_fcus->rght->v[i] != b_fcus->left)
	  {
	    Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);
	    l_infa  = 10.;
	    l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->rght->b[i]->l->v);
	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->rght->b[i],tree);
	  }

      Update_P_Lk(tree,b_fcus,b_fcus->rght);

      if(lk_temp < lk1 - tree->mod->s_opt->min_diff_lk_local)
	{
	  PhyML_Printf("\n== lk_temp = %f lk1 = %f\n",lk_temp,lk1);
	  PhyML_Printf("\n== Err in file %s at line %d\n\n",__FILE__,__LINE__);
	  Warn_And_Exit("");
	}
    }
  while(FABS(lk_temp-lk1) > tree->mod->s_opt->min_diff_lk_global);
  /*! until no significative improvement is detected */

  lk1 = tree->c_lnL;

  /*! save likelihood of each sites, in order to compute branch supports */

  buff = (t_tree *)tree;
  do
    {      
      counter=0;
      For(site,tree->n_pattern)
        {
          wei=0;
          For(wei,tree->data->wght[site])
            {
              tree->log_lks_aLRT[1][counter]= tree->c_lnL_sorted[site] / tree->data->wght[site];
              counter++;
            }
        }
      tree = tree->next_mixt;
    }
  while(tree);
  tree = (t_tree *)buff;

  /*! undo the swap */
  Swap(v3,b_fcus->left,b_fcus->rght,v2,tree);

  /*! Go back to initial branch lengths */
  MIXT_Set_Lengths_Of_This_Edge(len_e1,e1,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e2,e2,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e3,e3,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e4,e4,tree);
  MIXT_Set_Lengths_Of_This_Edge(bl_init,b_fcus,tree);

  Update_PMat_At_Given_Edge(e1,tree);
  Update_PMat_At_Given_Edge(e2,tree);
  Update_PMat_At_Given_Edge(e3,tree);
  Update_PMat_At_Given_Edge(e4,tree);
  Update_PMat_At_Given_Edge(b_fcus,tree);
  
  /* Sanity check */
  MIXT_Set_Alias_Subpatt(YES,tree);
  Update_P_Lk(tree,b_fcus,b_fcus->rght);
  Update_P_Lk(tree,b_fcus,b_fcus->left);
  lk_temp = Lk(b_fcus,tree);
  MIXT_Set_Alias_Subpatt(NO,tree);
  if((lk_temp > lk_init + tree->mod->s_opt->min_diff_lk_local) || (lk_temp < lk_init - tree->mod->s_opt->min_diff_lk_local))
    {
      PhyML_Printf("\n== lk_temp = %f lk_init = %f",lk_temp,lk_init);
      PhyML_Printf("\n== Err. in file %s at line %d\n",__FILE__,__LINE__);
      Warn_And_Exit("");
    }

  /*! do the second possible swap */
  Swap(v2,b_fcus->left,b_fcus->rght,v4,tree);

  Set_Both_Sides(YES,tree);
  MIXT_Set_Alias_Subpatt(YES,tree);
  Update_P_Lk(tree,b_fcus,b_fcus->left);
  Update_P_Lk(tree,b_fcus,b_fcus->rght);    
  lk2 = Lk(b_fcus,tree);
  MIXT_Set_Alias_Subpatt(NO,tree);

  For(i,3)
    if(b_fcus->left->v[i] != b_fcus->rght)
      Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);

  For(i,3)
    if(b_fcus->rght->v[i] != b_fcus->left)
      Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);

  /*! Optimize branch lengths and update likelihoods */
  lk_temp = UNLIKELY;
  do
    {
      lk2 = lk_temp;


      For(i,3)
	if(b_fcus->left->v[i] != b_fcus->rght)
	  {
	    Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);

	    l_infa  = 10.;
	    l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->left->b[i]->l->v);
	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->left->b[i],tree);

            if(lk_temp < lk2 - tree->mod->s_opt->min_diff_lk_local)
              {
                PhyML_Printf("\n== l_infa: %f l_infb: %f l: %f var:%f",l_infa,l_infb,b_fcus->left->b[i]->l->v,b_fcus->left->b[i]->l_var->v);
                PhyML_Printf("\n== lk_temp = %f lk2 = %f",lk_temp,lk2);
                PhyML_Printf("\n== Err. in file %s at line %d",__FILE__,__LINE__);
                Exit("\n");
              }
	  }

      Update_P_Lk(tree,b_fcus,b_fcus->left);

      l_infa  = 10.;
      l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->l->v);
      lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus,tree);

      if(lk_temp < lk2 - tree->mod->s_opt->min_diff_lk_local)
	{
          PhyML_Printf("\n== l_infa: %f l_infb: %f l: %f var:%f",l_infa,l_infb,b_fcus->l->v,b_fcus->l_var->v);
	  PhyML_Printf("\n== lk_temp = %f lk2 = %f",lk_temp,lk2);
	  PhyML_Printf("\n== Err. in file %s at line %d",__FILE__,__LINE__);
	  Exit("\n");
	}

      For(i,3)
	if(b_fcus->rght->v[i] != b_fcus->left)
	  {
	    Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);

                
	    l_infa  = 10.;
	    l_infb  = MAX(0.0,tree->mod->l_min/b_fcus->rght->b[i]->l->v);
	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->rght->b[i],tree);
            

            if(lk_temp < lk2 - tree->mod->s_opt->min_diff_lk_local)
              {
                Set_Both_Sides(YES,tree);
                Lk(b_fcus,tree);
                Check_Lk_At_Given_Edge(YES,tree);
                PhyML_Printf("\n== l_infa: %f l_infb: %f l: %f var:%f",l_infa,l_infb,b_fcus->rght->b[i]->l->v,b_fcus->rght->b[i]->l_var->v);
                PhyML_Printf("\n== lk_temp = %f lk2 = %f",lk_temp,lk2);
                PhyML_Printf("\n== Err. in file %s at line %d",__FILE__,__LINE__);
                Exit("\n");
              }
	  }

      Update_P_Lk(tree,b_fcus,b_fcus->rght);

    }
  while(FABS(lk_temp-lk2) > tree->mod->s_opt->min_diff_lk_global);
  //until no significative improvement is detected

  lk2 = tree->c_lnL;

  /*! save likelihood of each sites, in order to compute branch supports */
  buff = (t_tree *)tree;  
  do
    {
      counter=0;
      For(site,tree->n_pattern)
        {
          wei=0;
          For(wei,tree->data->wght[site])
            {
              tree->log_lks_aLRT[2][counter]= tree->c_lnL_sorted[site] / tree->data->wght[site];
              counter++;
            }
        }
      tree = tree->next_mixt;
    }
  while(tree);
  tree = (t_tree *)buff;


  /*! undo the swap */
  Swap(v4,b_fcus->left,b_fcus->rght,v2,tree);
  /***********/

  /*! restore the initial branch length values */
  MIXT_Set_Lengths_Of_This_Edge(len_e1,e1,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e2,e2,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e3,e3,tree);
  MIXT_Set_Lengths_Of_This_Edge(len_e4,e4,tree);
  MIXT_Set_Lengths_Of_This_Edge(bl_init,b_fcus,tree);

  /*! recompute likelihoods */
  Update_PMat_At_Given_Edge(e1,tree);
  Update_PMat_At_Given_Edge(e2,tree);
  Update_PMat_At_Given_Edge(e3,tree);
  Update_PMat_At_Given_Edge(e4,tree);
  Update_PMat_At_Given_Edge(b_fcus,tree);

  MIXT_Set_Alias_Subpatt(YES,tree);
  Update_P_Lk(tree,b_fcus,b_fcus->left);
  Update_P_Lk(tree,b_fcus,b_fcus->rght);

  For(i,3)
    if(b_fcus->left->v[i] != b_fcus->rght)
      Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);

  For(i,3)
    if(b_fcus->rght->v[i] != b_fcus->left)
      Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);

  MIXT_Set_Alias_Subpatt(NO,tree);

  lk_temp = Lk(b_fcus,tree);

  if((lk_temp > lk_init + tree->mod->s_opt->min_diff_lk_local) || (lk_temp < lk_init - tree->mod->s_opt->min_diff_lk_local))
    {
      PhyML_Printf("\n== lk_temp = %f lk_init = %f",lk_temp,lk_init);
      PhyML_Printf("\n== Err. in file %s at line %d\n",__FILE__,__LINE__);
      Warn_And_Exit("");
    }


  //save likelihoods in NNI structures
  b_fcus->nni->lk0 = lk0;
  b_fcus->nni->lk1 = lk1;
  b_fcus->nni->lk2 = lk2;

  b_fcus->nni->score = lk0 - MAX(lk1,lk2);

  if((b_fcus->nni->score <  tree->mod->s_opt->min_diff_lk_local) &&
     (b_fcus->nni->score > -tree->mod->s_opt->min_diff_lk_local))
    {
      b_fcus->nni->score = .0;
      b_fcus->nni->lk1 = b_fcus->nni->lk2 = b_fcus->nni->lk0;
    }

  if((b_fcus->nni->lk1 > b_fcus->nni->lk0) && (b_fcus->nni->lk1 > b_fcus->nni->lk2))
    {
      if(b_fcus->nni->lk0 > b_fcus->nni->lk2) result = 1; //lk1 > lk0 > lk2
      else                                    result = 3; //lk1 > lk2 > lk0
    }
  else if((b_fcus->nni->lk2 > b_fcus->nni->lk0) && (b_fcus->nni->lk2 > b_fcus->nni->lk1))
    {
      if(b_fcus->nni->lk0 > b_fcus->nni->lk1) result = 2; //lk2 > lk0 > lk1
      else                                    result = 4; //lk2 > lk1 > lk0
    }

  Free(len_e1);
  Free(len_e2);
  Free(len_e3);
  Free(len_e4);
  Free(bl_init);

  return result;
}

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/*
* Make one target swap, optimizing five branches.
* param tree : the tree to check
* param tested_t_edge : the swaping t_edge of the tree
* param swapToDo : 1 or 2, to select the first or the second swap to do
*/

void Make_Target_Swap(t_tree *tree, t_edge *b_fcus, int swaptodo)
{
  t_node *v1,*v2,*v3,*v4;
  phydbl lktodo;
  phydbl l_infa, l_infb;
  phydbl lk_init, lk_temp;
  int i;

  if(swaptodo < 0)
    {
      PhyML_Printf("\n== Err in file %s at line %d\n\n",__FILE__,__LINE__);
      Warn_And_Exit("");
    }

  //Initialization
  lk_init = tree->c_lnL;

  b_fcus->nni->score = .0;

  lktodo = UNLIKELY;
  v1 = v2 = v3 = v4 = NULL;

  v1 = b_fcus->left->v[b_fcus->l_v1];
  v2 = b_fcus->left->v[b_fcus->l_v2];
  v3 = b_fcus->rght->v[b_fcus->r_v1];
  v4 = b_fcus->rght->v[b_fcus->r_v2];

  if(v1->num < v2->num)
    {
      PhyML_Printf("\n== Err in file %s at line %d\n\n",__FILE__,__LINE__);
      Warn_And_Exit("");
    }
  if(v3->num < v4->num)
    {
      PhyML_Printf("\n. Err in file %s at line %d\n\n",__FILE__,__LINE__);
      Warn_And_Exit("");
    }


  /***********/
  //make the selected swap
  if(swaptodo==1)
    {
      Swap(v2,b_fcus->left,b_fcus->rght,v3,tree);
      if(!Check_Topo_Constraints(tree,tree->io->cstr_tree)) Swap(v3,b_fcus->left,b_fcus->rght,v2,tree);
    }
  else
    {
      Swap(v2,b_fcus->left,b_fcus->rght,v4,tree);
      if(!Check_Topo_Constraints(tree,tree->io->cstr_tree)) Swap(v4,b_fcus->left,b_fcus->rght,v2,tree);
    }

  MIXT_Set_Alias_Subpatt(YES,tree);
  Set_Both_Sides(YES,tree);     
  lktodo = Update_Lk_At_Given_Edge(b_fcus,tree);

  For(i,3)
    if(b_fcus->left->v[i] != b_fcus->rght)
      Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);

  For(i,3)
    if(b_fcus->rght->v[i] != b_fcus->left)
      Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);

  MIXT_Set_Alias_Subpatt(NO,tree);

  //Optimize branch lengths and update likelihoods
  lk_temp = UNLIKELY;
  do
    {
      lktodo = lk_temp;

      For(i,3)
	if(b_fcus->left->v[i] != b_fcus->rght)
	  {
	    Update_P_Lk(tree,b_fcus->left->b[i],b_fcus->left);

	    l_infa  = 10.;
	    l_infb  = tree->mod->l_min/b_fcus->left->b[i]->l->v;
	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->left->b[i],tree);
	  }


      Update_P_Lk(tree,b_fcus,b_fcus->left);

      l_infa  = 10.;
      l_infb  = tree->mod->l_min/b_fcus->l->v;
      lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus,tree);



      For(i,3)
	if(b_fcus->rght->v[i] != b_fcus->left)
	  {
	    Update_P_Lk(tree,b_fcus->rght->b[i],b_fcus->rght);

	    l_infa  = 10.;
	    l_infb  = tree->mod->l_min/b_fcus->rght->b[i]->l->v;
	    lk_temp = Br_Len_Brent(l_infb,l_infa,b_fcus->rght->b[i],tree);
	  }

      Update_P_Lk(tree,b_fcus,b_fcus->rght);

      if(lk_temp < lktodo - tree->mod->s_opt->min_diff_lk_local)
	{
	  PhyML_Printf("\n. Edge %3d lk_temp = %f lktodo = %f\n",b_fcus->num,lk_temp,lktodo);
	  PhyML_Printf("\n. Err in file %s at line %d\n\n",__FILE__,__LINE__);
	  Warn_And_Exit("");
	}
    }
  while(FABS(lk_temp-lktodo) > tree->mod->s_opt->min_diff_lk_global);
  //until no significative improvement is detected


/*   PhyML_Printf("\n.<< [%3d] l=%f lk_init=%f tree->c_lnL=%f score=%12f v1=%3d v2=%3d v3=%3d v4=%3d >>", */
/* 	 b_fcus->num, */
/* 	 b_fcus->l->v, */
/* 	 lk_init, */
/* 	 tree->c_lnL, */
/* 	 lk_init-tree->c_lnL, */
/* 	 v1->num,v2->num,v3->num,v4->num);       */


  if(tree->c_lnL < lk_init - tree->mod->s_opt->min_diff_lk_global)
    {
      PhyML_Printf("\n. [%3d] v1=%d v2=%d v3=%d v4=%d",
	     b_fcus->num,v1->num,v2->num,v3->num,v4->num);
      PhyML_Printf("\n. tree->c_lnL = %f lk_init = %f\n",tree->c_lnL,lk_init);
      PhyML_Printf("\n. Err in file %s at line %d\n\n",__FILE__,__LINE__);
      Warn_And_Exit("");
    }
}

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/**
* Convert an aLRT statistic to a none parametric support
* param in: the statistic
*/

phydbl Statistics_To_Probabilities(phydbl in)
{
  phydbl rough_value=0.0;
  phydbl a=0.0;
  phydbl b=0.0;
  phydbl fa=0.0;
  phydbl fb=0.0;

  if(in>=0.000000393 && in<0.00000157)
    {
      a=0.000000393;
      b=0.00000157;
      fa=0.0005;
      fb=0.001;
    }
  else if(in>=0.00000157 && in<0.0000393)
    {
      a=0.00000157;
      b=0.0000393;
      fa=0.001;
      fb=0.005;
    }
  else if(in>=0.0000393 && in<0.000157)
    {
      a=0.0000393;
      b=0.000157;
      fa=0.005;
      fb=0.01;
    }
  else if(in>=0.000157 && in<0.000982)
    {
      a=0.000157;
      b=0.000982;
      fa=0.01;
      fb=0.025;
    }
  else if(in>0.000982 && in<0.00393)
    {
      a=0.000982;
      b=0.00393;
      fa=0.025;
      fb=0.05;
    }
  else if(in>=0.00393 && in<0.0158)
    {
      a=0.00393;
      b=0.0158;
      fa=0.05;
      fb=0.1;
    }
  else if(in>=0.0158 && in<0.0642)
    {
      a=0.0158;
      b=0.0642;
      fa=0.1;
      fb=0.2;
    }
  else if(in>=0.0642 && in<0.148)
    {
      a=0.0642;
      b=0.148;
      fa=0.2;
      fb=0.3;
    }
  else if(in>=0.148 && in<0.275)
    {
      a=0.148;
      b=0.275;
      fa=0.3;
      fb=0.4;
    }
  else if(in>=0.275 && in<0.455)
    {
      a=0.275;
      b=0.455;
      fa=0.4;
      fb=0.5;
    }
  else if(in>=0.455 && in<0.708)
    {
      a=0.455;
      b=0.708;
      fa=0.5;
      fb=0.6;
    }
  else if(in>=0.708 && in<1.074)
    {
      a=0.708;
      b=1.074;
      fa=0.6;
      fb=0.7;
    }
  else if(in>=1.074 && in<1.642)
    {
      a=1.074;
      b=1.642;
      fa=0.7;
      fb=0.8;
    }
  else if(in>=1.642 && in<2.706)
    {
      a=1.642;
      b=2.706;
      fa=0.8;
      fb=0.9;
    }
  else if(in>=2.706 && in<3.841)
    {
      a=2.706;
      b=3.841;
      fa=0.9;
      fb=0.95;
    }
  else if(in>=3.841 && in<5.024)
    {
      a=3.841;
      b=5.024;
      fa=0.95;
      fb=0.975;
    }
  else if(in>=5.024 && in<6.635)
    {
      a=5.024;
      b=6.635;
      fa=0.975;
      fb=0.99;
    }
  else if(in>=6.635 && in<7.879)
    {
      a=6.635;
      b=7.879;
      fa=0.99;
      fb=0.995;
    }
  else if(in>=7.879 && in<10.828)
    {
      a=7.879;
      b=10.828;
      fa=0.995;
      fb=0.999;
    }
  else if(in>=10.828 && in<12.116)
    {
      a=10.828;
      b=12.116;
      fa=0.999;
      fb=0.9995;
    }
  if (in>=12.116)
    {
      rough_value=0.9999;
    }
  else if(in<0.000000393)
    {
      rough_value=0.0001;
    }
  else
    {
      rough_value=(b-in)/(b-a)*fa + (in - a)/(b-a)*fb;
    }
  rough_value=rough_value+(1.0-rough_value)/2.0;
  rough_value=rough_value*rough_value*rough_value;
  return rough_value;
}

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/**
* deprecated
* Compute a RELL support, using the latest tested branch
* param tree: the tested tree
*/
phydbl Statistics_to_RELL(t_tree *tree)
{
  int i;
  int occurence=1000;
  phydbl nb=0.0;
  phydbl res;
  int site;
  phydbl lk0=0.0;
  phydbl lk1=0.0;
  phydbl lk2=0.0;
  phydbl buff = -1.;
  int position = -1;
  t_tree *buff_tree;

  /*! 1000 times */
  For(i,occurence)
    {
      lk0=0.0;
      lk1=0.0;
      lk2=0.0;

      /*! Shuffle the data and increment the support, if needed */
      buff_tree = tree;
      do
        {
          For(site, tree->data->init_len)
            {
              buff  = rand();
              buff /= (RAND_MAX+1.);
              buff *= tree->data->init_len;
              position = (int)FLOOR(buff);
              
              lk0+=tree->log_lks_aLRT[0][position];
              lk1+=tree->log_lks_aLRT[1][position];
              lk2+=tree->log_lks_aLRT[2][position];
            }
          if (lk0>=lk1 && lk0>=lk2) nb++;
          tree = tree->next_mixt;
        }
      while(tree);
      tree = buff_tree;
    }

  res= nb/(phydbl)occurence;

  return res;
}
//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/*!
  Compute a SH-like support, using the latest tested branch
  param tree: the tested tree
*/
phydbl Statistics_To_SH(t_tree *tree)
{
  int i;
  int occurence=1000;
  phydbl nb=0.0;
  phydbl res;
  int site;
  phydbl lk0=0.0;
  phydbl lk1=0.0;
  phydbl lk2=0.0;
  phydbl c0=0.0;
  phydbl c1=0.0;
  phydbl c2=0.0;
  phydbl buff=-1.;
  int position=-1;
  phydbl delta_local=-1.;
  phydbl delta=0.0;
  t_tree *buff_tree;


  /*! Compute the total log-lk of each NNI position */
  buff_tree = tree;
  do
    {
      For(site, tree->data->init_len)
        {
          c0+=tree->log_lks_aLRT[0][site];
          c1+=tree->log_lks_aLRT[1][site];
          c2+=tree->log_lks_aLRT[2][site];
        }
      tree = tree->next_mixt;
    }
  while(tree);
  tree = buff_tree;

  
  if (c0>=c1 && c0>=c2)
    {
      if (c1>=c2)
        {
          delta=c0-c1;
        }
      else
        {
          delta=c0-c2;
        }
    }
  else if(c1>=c0 && c1>=c2)
    {
      if (c0>=c2)
        {
          delta=c1-c0;
        }
      else
        {
          delta=c1-c2;
        }
    }
  else
    {
      if (c1>=c0)        
        {
          delta=c2-c1;
        }
      else
        {
          delta=c2-c0;
        }
    }

  /*! 1000 times */
  For(i,occurence)
    {
      lk0=0.0;
      lk1=0.0;
      lk2=0.0;
      
      buff_tree = tree;
      do
        {
          /*! Shuffle the data */
          For(site, tree->data->init_len)
            {
              buff  = rand();
              buff /= (RAND_MAX+1.);
              buff *= tree->data->init_len;
              position = (int)FLOOR(buff);
              
              lk0+=tree->log_lks_aLRT[0][position];
              lk1+=tree->log_lks_aLRT[1][position];
              lk2+=tree->log_lks_aLRT[2][position];
            }
          
          tree = tree->next_mixt;
        }
      while(tree);
      tree = buff_tree;

      /*! return to null hypothesis */
      lk0=lk0-c0;
      lk1=lk1-c1;
      lk2=lk2-c2;
      
      /*! compute results and increment if needed */
      delta_local=0.0;
      if (lk0>=lk1 && lk0>=lk2)
	{
	  if (lk1>=lk2)
	    {
	      delta_local=lk0-lk1;
	    }
	  else
	    {
	      delta_local=lk0-lk2;
	    }
	}
      else if(lk1>=lk0 && lk1>=lk2)
	{
	  if (lk0>=lk2)
	    {
	      delta_local=lk1-lk0;
	    }
	  else
	    {
	      delta_local=lk1-lk2;
	    }
	}
      else
	{
	  if (lk1>=lk0)
            
	    {
	      delta_local=lk2-lk1;
	    }
	  else
	    {
	      delta_local=lk2-lk0;
	    }
	}
      
      if (delta>(delta_local+0.1)) 
        {
          nb++;
        }
    }
  
  res= nb/occurence;
  
  return res;
}

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

/**
* deprecated
* Compute one side likelihood
* param b_fcus : concerned edge
* param tree : b_fcus tree
* param exclude :  side to exclude for computation
*/
phydbl Update_Lk_At_Given_Edge_Excluding(t_edge *b_fcus, t_tree *tree, t_node *exclude)
{
  if((!b_fcus->left->tax) && (exclude == NULL || exclude != b_fcus->left))
    Update_P_Lk(tree,b_fcus,b_fcus->left);
  if((!b_fcus->rght->tax) && (exclude == NULL || exclude != b_fcus->rght))
    Update_P_Lk(tree,b_fcus,b_fcus->rght);

  tree->c_lnL = Lk(b_fcus,tree);

  return tree->c_lnL;
}


//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////

//////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////