diff --git a/lib/lbfgs.c b/lib/lbfgs.c
index e9e587b..ee638be 100644
--- a/lib/lbfgs.c
+++ b/lib/lbfgs.c
@@ -169,6 +169,15 @@ static int update_trial_interval(
     int *brackt
     );
 
+static lbfgsfloatval_t orthantwise_gnorm(
+    const lbfgsfloatval_t* x,
+    const lbfgsfloatval_t* g,
+    const lbfgsfloatval_t c,
+    const int start,
+    const int n
+    );
+
+
 #if     defined(USE_SSE) && (defined(__SSE__) || defined(__SSE2__))
 static int round_out_variables(int n)
 {
@@ -357,8 +366,12 @@ int lbfgs(
     /*
        Make sure that the initial variables are not a minimizer.
      */
-    vecnorm(&gnorm, g, n);
     vecnorm(&xnorm, x, n);
+    if (param->orthantwise_c != 0.) {
+        gnorm = orthantwise_gnorm(x, g, param->orthantwise_c, param->orthantwise_start, n);
+    } else {
+        vecnorm(&gnorm, g, n);
+    }
     if (xnorm < 1.0) xnorm = 1.0;
     if (gnorm / xnorm <= param->epsilon) {
         ret = LBFGS_ALREADY_MINIMIZED;
@@ -385,8 +398,12 @@ int lbfgs(
         }
 
         /* Compute x and g norms. */
-        vecnorm(&gnorm, g, n);
         vecnorm(&xnorm, x, n);
+        if (param->orthantwise_c != 0.) {
+            gnorm = orthantwise_gnorm(x, g, param->orthantwise_c, param->orthantwise_start, n);
+        } else {
+            vecnorm(&gnorm, g, n);
+        }
 
         /* Report the progress. */
         if (cd.proc_progress) {
@@ -1217,3 +1234,32 @@ static int update_trial_interval(
     *t = newt;
     return 0;
 }
+
+static lbfgsfloatval_t orthantwise_gnorm(
+    const lbfgsfloatval_t* x,
+    const lbfgsfloatval_t* g,
+    const lbfgsfloatval_t c,
+    const int start,
+    const int n
+    )
+{
+    int i;
+    lbfgsfloatval_t d = 0.;
+    lbfgsfloatval_t norm = 0.;
+
+    for (i = 0;i < start;++i) {
+        norm += g[i] * g[i];
+    }
+
+    for (i = start;i < n;++i) {
+        d = g[i];
+        if (x[i] < 0.) {
+            d -= c;
+        } else if (0. < x[i]) {
+            d += c;
+        }
+        norm += d * d;
+    }
+
+    return sqrt(norm);
+}