Demystifying Midgetman MGM-134A
Alexander Levakov, Senior Research Fellow, Ph.D
November 30, 2018
Midgetman MGM-134A
Midgetman
The MGM-134A Midgetman, also known as the Small Intercontinental Ballistic Missile (SICBM), was an intercontinental ballistic missile developed by the United States Air Force. The system was mobile and could be set up rapidly, allowing it to move to a new firing location after learning of an enemy missile launch. To attack the weapon, the enemy would have to blanket the area around its last known location with multiple warheads, using up a large percentage of their force for limited gains and no guarantee that all of the missiles would be destroyed. In such a scenario, the U.S. would retain enough of their forces for a successful counterstrike, thereby maintaining a deterrence.
For more see: https://en.wikipedia.org/wiki/MGM-134_Midgetman
SLBM data
load("slbm.dat")
library(DT)
datatable(slbm)library(ggplot2)
library(GGally)
#Here we use function from https://www.r-bloggers.com/multiple-regression-lines-in-ggpairs/
my_fn <- function(data, mapping, ...){
p <- ggplot(data = data, mapping = mapping) +
geom_point() +
geom_smooth(method=loess, fill="red", color="red", ...) +
geom_smooth(method=lm, fill="blue", color="blue", ...)
p
}
g = ggpairs(slbm,columns = 2:6, lower = list(continuous = my_fn))
g
Bayesian Linear Regression Model
library(rstanarm)## Loading required package: Rcpp## rstanarm (Version 2.17.3, packaged: 2018-02-17 05:11:16 UTC)## - Do not expect the default priors to remain the same in future rstanarm versions.## Thus, R scripts should specify priors explicitly, even if they are just the defaults.## - For execution on a local, multicore CPU with excess RAM we recommend calling## options(mc.cores = parallel::detectCores())## - Plotting theme set to bayesplot::theme_default().library(bayesplot)## This is bayesplot version 1.4.0## - Plotting theme set to bayesplot::theme_default()## - Online documentation at mc-stan.org/bayesplotoptions(mc.cores = parallel::detectCores())
fit.slbm.bs<-stan_glm(sqrt(R)~S+D+L+W+log(M)+log(P),
data=slbm,chains=4,iter=10000,seed=12345)
fit.slbm.bs## stan_glm
## family: gaussian [identity]
## formula: sqrt(R) ~ S + D + L + W + log(M) + log(P)
## observations: 26
## predictors: 7
## ------
## Median MAD_SD
## (Intercept) -40.6 106.7
## S 9.4 5.1
## D 37.4 19.8
## L 0.9 1.7
## W 3.3 5.6
## log(M) 6.5 15.0
## log(P) -7.3 6.0
## sigma 10.2 1.7
##
## Sample avg. posterior predictive distribution of y:
## Median MAD_SD
## mean_PPD 69.0 2.8
##
## ------
## For info on the priors used see help('prior_summary.stanreg').plot(fit.slbm.bs)
posterior_vs_prior(fit.slbm.bs)##
## Drawing from prior...
fit.slbm.bs2<-as.array(fit.slbm.bs)
mcmc_hist(fit.slbm.bs2)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
mcmc_trace(fit.slbm.bs2)
summary(fit.slbm.bs)##
## Model Info:
##
## function: stan_glm
## family: gaussian [identity]
## formula: sqrt(R) ~ S + D + L + W + log(M) + log(P)
## algorithm: sampling
## priors: see help('prior_summary')
## sample: 20000 (posterior sample size)
## observations: 26
## predictors: 7
##
## Estimates:
## mean sd 2.5% 25% 50% 75% 97.5%
## (Intercept) -40.0 109.1 -254.9 -112.3 -40.6 31.7 175.7
## S 9.4 5.4 -1.2 5.9 9.4 12.9 20.2
## D 37.4 20.3 -2.6 24.1 37.4 50.9 77.4
## L 0.9 1.8 -2.7 -0.3 0.9 2.0 4.4
## W 3.2 5.8 -8.5 -0.5 3.3 7.1 14.8
## log(M) 6.4 15.6 -24.3 -3.6 6.5 16.6 37.0
## log(P) -7.3 6.2 -19.5 -11.3 -7.3 -3.3 5.0
## sigma 10.4 1.8 7.6 9.2 10.2 11.4 14.6
## mean_PPD 69.0 2.9 63.2 67.1 69.0 70.9 74.7
## log-posterior -110.5 2.4 -116.3 -111.8 -110.1 -108.8 -107.1
##
## Diagnostics:
## mcse Rhat n_eff
## (Intercept) 1.2 1.0 8809
## S 0.0 1.0 12972
## D 0.2 1.0 9063
## L 0.0 1.0 12118
## W 0.1 1.0 11991
## log(M) 0.2 1.0 8316
## log(P) 0.1 1.0 13099
## sigma 0.0 1.0 10055
## mean_PPD 0.0 1.0 19069
## log-posterior 0.0 1.0 6234
##
## For each parameter, mcse is Monte Carlo standard error, n_eff is a crude measure of effective sample size, and Rhat is the potential scale reduction factor on split chains (at convergence Rhat=1).MGM-134A Operational Range
MGM-134A Mobile Launcher
slbm.pp <- posterior_predict(fit.slbm.bs,
newdata = data.frame(L=14,D=1.17,S=3,W=1,
M=13600,P=600),seed=12345)
#Note. Payload is an expert estimate for the model
Range.km <- slbm.pp^2
Range.km <- Range.km[1:10000,]
quantile(Range.km,probs = c(0.05,0.5,0.95))## 5% 50% 95%
## 1210.197 3812.865 7847.878Range.km <- slbm.pp^2
Range.km <- Range.km[1:10000,]
ggplot(data=as.data.frame(Range.km), aes(Range.km)) +
geom_histogram(bins = 30,col="black",fill="green") +
geom_vline(xintercept = mean(Range.km), color = "red") +
geom_errorbarh(aes(y=-5, xmin=quantile(Range.km,0.1),
xmax=quantile(Range.km,0.9)),
data=as.data.frame(Range.km), col="#0094EA", size=3) +
ggtitle(label="Probability density of MGM-134A range",
subtitle = "P(1210<Range.km<7848)=0.9")
Conclusions
1.Applying Bayesian linear regression model for SLBM data (M,P,D,L,S,W) we can produce posterior distribution sample given MGM-134A data (M=13600 kg,P=600 kg,D=1.17 m,L=14 m,S=3,W=1) with 90% credible interval \(P(1210\le Range\le7848)=0.9\) and the mean \(Range=3812\).
2.MGM-134A SLBM had operational Range corresponding to ICBM which could survive bolt-out-of-the blue attack and hit any possible target world wide.