Convergence issues

[Asier García Escarzaga]

Dear all, 

Multiple runs of the model below result in widly different estimates. I have tried to improve consistency by changing resolution and increasing kIterations (up to 3000) but the problem persists. Any suggestions on how to address this?

Thanks in advance, 

Asier

 Options()
 {
  Resolution=5;
  kIterations=3000;
 };
 Plot()
 {
  Outlier_Model("General",T(5),U(0,4),"t");
  Curve("IntCal20","IntCal20.14c");
  Curve("Marine20","Marine20.14c");
  Sequence ("General")
  {
   Phase ("Phase 102")
   {
    Boundary ("Start Phase 102");
    Phase("Unit 102")
    {
     Curve("=IntCal20");
     R_Date("OxA-40820",15542,65)
     {
      Outlier("General", 0.05);
     };
     Interval("Duration of Unit 102");
    };
   };
   Boundary ("End Phase 102");
   Boundary ("Start Phase Pits");
   Phase ("Pits")
   {
    Sequence("Pit 108")
    {
     Boundary("Start Unit 108");
     Phase("Unit 108")
     {
      Curve("=IntCal20");
      R_Date("OxA-43812", 14954, 51)
      {
       Outlier("General", 0.05);
      };
      Curve("=Marine20");
      Delta_R("Littorina_littorea",U(-800,800));
      R_Date("OxA-44338",15680,34)
      {
       Outlier("General", 0.05);
      };
      R_Date("OxA-44337",15374,34)
      {
       Outlier("General", 0.05);
      };
      Interval("Duration of Unit 108");
     };
     Boundary("End Unit 108");
    };
    Sequence("Pit 106")
    {
     Boundary("Start Unit 106low");
     Phase("Unit 106low")
     {
      Curve("=IntCal20");
      R_Date("OxA-43870", 14906, 58)
      {
       Outlier("General", 0.05);
      };
      Curve("=Marine20");
      Delta_R("=Littorina_littorea");
      R_Date("OxA-40870",15375,28)
      {
       Outlier("General", 0.05);
      };
      R_Date("OxA-43669", 15488,29)
      {
       Outlier("General", 0.05);
      };
      Interval("Duration of Unit 106low");
     };
     Boundary("End Unit 106low");
     Boundary("Start Unit 106up");
     Phase("Unit 106up")
     {
      Curve("=IntCal20");
      R_Date("OxA-43811",14942, 51)
      {
       Outlier("General", 0.05);
      };
      Curve("=Marine20");
      Delta_R("=Littorina_littorea");
      R_Date("OxA-43872",15488,36)
      {
       Outlier("General", 0.05);
      };
      R_Date("OxA-43871",15489,35)
      {
       Outlier("General", 0.05);
      };
      Interval("Duration of Unit 106up");
     };
     Boundary("End Unit 106up");
    };
    Sequence("Pit 107")
    {
     Boundary("Start Unit 107");
     Phase("Unit 107")
     {
      Curve("=IntCal20");
      R_Date("OxA-43813", 14787, 57)
      {
       Outlier("General", 0.05);
      };
      Curve("=Marine20");
      Delta_R("=Littorina_littorea");
      R_Date("OxA-44340",15302,33)
      {
       Outlier("General", 0.05);
      };
      R_Date("OxA-44339",15468,33)
      {
       Outlier("General", 0.05);
      };
      Interval("Duration of Unit 107");
     };
     Boundary("End Unit 107");
    };
   };
   Boundary ("End Phase Pits");
   Boundary("Start Unit 209");
   Phase("Unit 209")
   {
    Curve("=IntCal20");
    R_Date("OxA-3112-25",15100,67)
    {
     Outlier("General", 0.05);
    };
    R_Date("OxA-41023",15121,56)
    {
     Outlier("General", 0.05);
    };
    Curve("=Marine20");
    Delta_R("=Littorina_littorea");
    R_Date("OxA-44341",15670,34)
    {
     Outlier("General", 0.05);
    };
    Delta_R("Patella_vulgata",U(-800,800));
    R_Date("OxA-44342",15473,33)
    {
     Outlier("General", 0.05);
    };
    Interval("Duration of Unit 209");
   };
   Boundary("End Unit 209");
  };
 };

[Christopher Ramsey]

Dear Asier

I think the first boundary is in the wrong place and should be:

Options()
{
Resolution=5;
};

Plot()
{
Outlier_Model("General",T(5),U(0,4),"t");
Curve("IntCal20","IntCal20.14c");
Curve("Marine20","Marine20.14c");
Sequence ("General")
{

The convergence is still slow but I think this will converge given time. The reason that it is slow is that you have three complications in this model;

1. Outlier analysis - and in particular for phase 102 there is only one date. Whether this date is correct or not radically changes the first boundary and to some extent the rest of the model because it is much earlier.

2. Marine Delta_R which always slows convergence somewhat.

3. Nested phase groups - again these slow things down.

The model as above without the first phase seems to converge faster suggesting that is the main issue issue here.

Best wishes

Christopher

> --
> You received this message because you are subscribed to the Google Groups "OxCal" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to oxcal+un...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/oxcal/cb28c667-2f15-4679-a7bc-ae742dca4474n%40googlegroups.com.