I changed my reference group for a predictor and now my model won't run

Question

I have a bit of a strange question. I ran the following model, which includes as one of the predictors 'Valence.c'. This is predictor coded as '0' or '1', representing 'positive' and 'negative'. The predictor was centered so is actually '-0.5'and '0.5'.

> loss.1 <- glmer.nb(Loss_across.Chain ~ Posn.c*Valence.c + (Valence.c|mood.c/Chain), data = FinalData_forpoisson, control = glmerControl(optimizer = "bobyqa", check.conv.grad = .makeCC("warning", 0.05)))

I got the following output:

Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) ['glmerMod']
 Family: Negative Binomial(4.9852)  ( log )
Formula: Loss_across.Chain ~ Posn.c * Valence.c + (Valence.c | mood.c/Chain)
   Data: FinalData_forpoisson
Control: ..3

     AIC      BIC   logLik deviance df.resid 
  1894.7   1945.3   -936.4   1872.7      725 

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-1.3882 -0.7225 -0.5190  0.4375  7.1873 

Random effects:
 Groups       Name        Variance  Std.Dev.  Corr
 Chain:mood.c (Intercept) 8.782e-15 9.371e-08     
              Valence.c   9.608e-15 9.802e-08 0.48
 mood.c       (Intercept) 0.000e+00 0.000e+00     
              Valence.c   1.654e-14 1.286e-07  NaN
Number of obs: 736, groups:  Chain:mood.c, 92; mood.c, 2

Fixed effects:
                 Estimate Std. Error z value Pr(>|z|)    
(Intercept)      -0.19255    0.04794  -4.016 5.92e-05 ***
Posn.c           -0.61011    0.04122 -14.800  < 2e-16 ***
Valence.c        -0.27372    0.09589  -2.855  0.00431 ** 
Posn.c:Valence.c  0.38043    0.08245   4.614 3.95e-06 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Correlation of Fixed Effects:
            (Intr) Posn.c Vlnc.c
Posn.c       0.491              
Valence.c    0.029 -0.090       
Psn.c:Vlnc. -0.090  0.062  0.491

As the fixed effect for Valence.c was negative I thought I would try re-code the variable so that positive was now '0.5' and negative was now '-0.5'. I thought explaining an increase in the incident rate would be easier than explaining a decrease. So I ran this model which is the same, except the datafile it calls has the reverse codings:

> loss.2 <- glmer.nb(Loss_across.Chain ~ Posn.c*Valence.c + (Valence.c|mood.c/Chain), data = LossAnalysis_ValenceCodingReversed, control = glmerControl(optimizer = "bobyqa", check.conv.grad = .makeCC("warning", 0.05)))

I got this warning message:

Warning messages:
1: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  unable to evaluate scaled gradient
2: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model failed to converge: degenerate  Hessian with 1 negative eigenvalues

Why would changing the reference group mean that the model now fails to converge?? I have the same number of observations for positive and negative. Any help would be great!

Thanks