Skygrid throws Newton-Raphson error even with reasonable starting values

[maxbiostat]

if one runs this XML with beast-dev -seed 666 -threads 1 one gets

# BEAST v1.9.0pre20170207 Prerelease #20fe946
# Generated Wed Jun 14 12:12:38 BST 2017 [seed=666]
# -beagle_SSE -seed 666 -threads 1 Dengue4.env_vanilla.xml
state   Posterior       Prior           Likelihood      rootAge         ucld.mean   
0   -22178.9502     -1440.1547      -20738.7955     1936.00         1.00000         -
1000    -8567.2802      -405.2073       -8162.0729      1947.06         0.27465         -
2000    -5799.6373      -367.5387       -5432.0986      1953.91         5.23059E-2      -
Exception in thread "Thread-1" java.lang.RuntimeException: Newton Raphson algorithm did not converge within 200 step to a norm less than 0.01
Try starting BEAST with a more accurate initial tree.
    at dr.evomodel.coalescent.operators.GMRFMultilocusSkyrideBlockUpdateOperator.newNewtonRaphson(Unknown Source)
    at dr.evomodel.coalescent.operators.GMRFMultilocusSkyrideBlockUpdateOperator.newtonRaphson(Unknown Source)
    at dr.evomodel.coalescent.operators.GMRFMultilocusSkyrideBlockUpdateOperator.doOperation(Unknown Source)
    at dr.inference.operators.SimpleMCMCOperator.operate(Unknown Source)
    at dr.inference.markovchain.MarkovChain.runChain(Unknown Source)
    at dr.inference.mcmc.MCMC.chain(Unknown Source)
    at dr.inference.mcmc.MCMC.run(Unknown Source)
    at java.lang.Thread.run(Thread.java:748)
2.934 seconds

This is connected to #875, #888 and #746 but imho somewhat different because starting values for logPopSizes are reasonable.

[maxbiostat]

To reproduce this error, one must do wget https://github.com/beast-dev/beast-mcmc/archive/46ca0687ef9ede34cafeca8721c1fe52b0e95547.zip, extract, and compile (ant). This gets a snapshot of the repo roundabout the time this was reported.

[rambaut]

Related to #1030 - closing