=== Init ===

Run long enough to see good Higgs correlators at several values of the cutoff.

Two values of lambda, 1e-4 and 1e-6.

Four values of the bare vev: 0.1, 0.2, 0.5, and 1.0

Run massively on kaon: 5*4 runs for each (lambda,vev).


=== Runtime ===

Measure while updating ... or not: measurements were screwed; use cfgs (wrote
every 20th, have 2520 cfgs for each run) for measurement.  Run 
higgsmass/bin/corr on kaon head node.  Analyze in lambda=*/v=*/analysis;
link all cfgs together in ../analysis/allcfgs.

Copy analysis output to xeon for ising calculations.


=== Analysis ===

* Ising2_run subtracts <bar{phi}^2> (i.e. per config), which has a "negative
  norm", with which ising2_run does not deal well.
  
* Analyzing naively, subtracting <bar{phi}>^2, gives a good cosh, but with
  very large errors.
  
* Analyzing naively, subtracting <bar{phi}^2>, gives very small errors *and* a
  great cosh, using gnuplot to fit the form: A + B*cosh(m (T-t)) !  ising2_run
  does not seem to have the capacity to add in the A.  The histories look
  great, and the points are separated by 20 Fourier accelerated updates, so I
  insist that this result is completely acceptable.  The only down side is the
  fit parameter correlation, the exponent with the coefficient.
  
* Subtracting <bar{phi}>^2 by hand and using ising2_run gives okay plateaux,
  but with unacceptably large errors (as in the naive analysis, as expected
  for good data).
  
* Using Dani's jacknife analysis, which subtracts <bar{phi}^2>, with the A 
  added, gives roughly the same excellent result as the naive analysis.  The
  errors are larger, coming from solving
      c_boot(t)/c_booot(t+1) = cosh(m_boot t)/cosh(m_booot (t+1))
  for each booststrap sample.  Dani claims that his way is the standard.
  After much consideration and comparison, I am willing to accept it, but with
  the notion that it may overestimate the errors by a factor of around 2.

* kaon side is done.  Have p=(0,0,0).bin, vev.bin, and higgs_propagator.bin
  for every run (in analysis directories).

* m_R errors are larger than m_phys ones; should not be the case.  Use more
  momenta, up to 8 (5k pts).


=== Results ===

* First, vevs are roughly 0.20, 0.23, 0.39, and 0.85 for lambda=1e-4 and
  0.21, 0.25, 0.47, and 1.03 for lambda=1e-6.  Notice that the vev doesn't
  change much from "v=0.2" to "v=0.1" in either case, a sign of finite volume
  effects.

* Using Dani's method for the effective mass gives decent results for v=0.5,1.0
  -- i.e., O(1%) errors and pretty good m_eff plots; but for v=0.1,0.2 the m_eff
  plots are pretty shoddy.  Still, I think it safe to say we have an O(5%)
  determination of m_eff, with possibly large finite volume effects.

* Agreement between m_corr (m_phys) and m_prop (m_R) is okay; errors are
  comparable.

* Mass does not decrease with decreasing vev except for v=0.5,1.0.  Finite
? volume effect on v=0.1,0.2?  The effect is much less for m_prop than m_corr.
  Recall (from a few lines up) that the vev doesn't change much from "v=0.2"
  to "v=0.1", so 1) the mass need not change much either and 2) the volume is
  not large enough.
...

NOT FINISHED